Cognitive Affective amp Behavioral Neuroscience2003 3 (4) 323-334

Our memories of past experiences are more fragilethan many people realize Not only are they vulnerable topoor encoding and forgetting but memory retrieval isalso highly susceptible to distortion (see eg Loftus ampLoftus 1980 Schacter 1999) These distorted memo-ries or false memories are recollections that are inaccu-rate misattributed or ldquorecollectionsrdquo of entirely novelevents (see eg Loftus amp Loftus 1980 Roediger amp Mc-Dermott 1995 Schacter 1999) Some false memoriescan be so influential and convincing that individuals be-lieve they have experienced a particular event that is en-tirely novel (Loftus amp Pickrell 1995) False memoriescan occur because memories are not exact reproductionsof past events (see eg Roediger 1996 Schacter Nor-man amp Koutstaal 1998) Rather they are complex re-constructions easily interfered with and heavily influencedby previously formed memories and mental representa-tions (see eg Loftus Miller amp Burns 1978 Roedigeramp McDermott 1995 Schacter et al 1998)

This reconstructive process makes memories vulnerableto misattribution Misattribution occurs when a particularmemory is attributed to an incorrect time place or person(Roediger 1996 Schacter 1999) The DeeseRoedigerndashMcDermott (DRM) paradigm initially developed byDeese (1959) and substantially refined by Roediger andMcDermott (1995) is frequently used to examine falserecognition since it is a very robust and reliable means ofinducing false memories in participants In this para-

digm participants study lists of semantically associatedwords but do not study a critical related lure word Insubsequent recall or recognition tests the participantswill not only frequently endorse as studied these criticallure words but will also often freely recall these unstud-ied critical items (Roediger amp McDermott 1995) Oneexplanation for this phenomenon is that participants areactually generating their own nonpresented lure wordduring the study phase When asked to recall words theyconfuse whether the word had actually been presented orthey themselves had conjured it up (Roediger amp McDer-mott 1995 Schacter et al 1998)

Marcia Johnsonrsquos source monitoring framework (SMF)suggests that during retrieval memories are reconstructeddifferently from the way they were initially formed en-abling various errors to be introduced such as features ofsimilar events or imagined events that activate together toform an inaccurate memory Thus evaluation of the mem-ory through monitoring processes ranging from automaticto deliberate is critical in determining the veridicalityand source accuracy of the recollected memory (see egJohnson Hashtroudi amp Lindsay 1993 Johnson amp Raye2000) Similarly Schacter et alrsquos (1998) constructivememory framework (CMF) explains how memory re-construction involves focusing on specific aspects of amemory so that irrelevant information is not retrievedand setting criteria to determine whether a memory hassuff icient perceptual vividness semantic detail andother phenomenal characteristics to be accurate

Often reality monitoring is used to discriminate mem-ories that are generated by reflection or imagination fromthose based on perception Externally perceived infor-mation and internally generated information may havesimilar semantic and perceptual content and may be pro-cessed similarly although externally perceived memo-ries typically have more of the information as well as

323 Copyright 2003 Psychonomic Society Inc

The research presented in this article was funded by Johns HopkinsUniversity and NSF Grant BCS-0236431 We thank LisaCaitlin PerriMarcel Poisot and the staff of the F M Kirby Center for FunctionalBrain Imaging for their assistance in data collection Correspondenceconcerning this article should be addressed to C Stark Department ofPsychological and Brain Sciences Johns Hopkins University Ames Hall3400 N Charles St Baltimore MD 21218 (e-mail cstarkjhuedu)

Neural processing associated with true and false memory retrieval

YOKO OKADO and CRAIG STARKJohns Hopkins University Baltimore Maryland

We investigated the neural bases for false memory with fMRI by examining neural activity during re-trieval processes that yielded true or false memories We used a reality monitoring paradigm in whichparticipants saw or imagined pictures of concrete objects (A subsequent misinformation task was alsoused to increase false memory rates) At test fMRI data were collected as the participants determinedwhether they had seen or had only imagined the object at study True memories were of seen picturesaccurately endorsed as seen and for false memories were of imagined pictures falsely endorsed as seenThree distinct patterns of activity were observed Left frontal and parietal activity was not different fortrue and for false memories whereas activity was greater for true than for false memories in occipitalvisual regions and posterior portions of the parahippocampal gyrus and activity was greater for falsethan for true memories in right anterior cingulate gyrus Possible interpretations are discussed

324 OKADO AND STARK

contextual detail (Johnson 1997 Johnson amp Raye 1981)Reality monitoring paradigms are used to investigatehow people discriminate externally generated and inter-nally generated memories Behavioral studies show thatfailures in reality monitoring can occur when events thatwere imagined are remembered with considerable sen-sory detail and mistakenly believed to have actually oc-curred (eg Johnson amp Raye 1981 Johnson Raye Wangamp Taylor 1979 Johnson Taylor amp Raye 1977) Be-cause the brain mechanisms involved in mental imagerymay be similar to those used in perceiving (see egFarah 1989 Kosslyn et al 1999 Wheeler Petersen ampBuckner 2000) it is conceivable that such an overlapcan cause imagined events to be misperceived as real

The propensity of misattribution can be increased bysuggestion Loftus and colleagues discovered that whenpeople are given suggestive and misleading informationabout a previous event their recollections of the originalevent can be altered by the misinformation (eg Loftuset al 1978 Loftus amp Palmer 1974) It has been sug-gested that the misinformation may be replacing theoriginal memory (Loftus et al 1978) or it may be pro-viding information that competes with the original mem-ory during retrieval (McCloskey amp Zaragoza 1985) Re-gardless it is clear that misinformation decreases thelikelihood of accurate memory retrieval and increasesthe chances that misinformation will be misattributed tothe original event

Although behavioral studies have extensively investi-gated and provided evidence for the existence and cre-ation of false memories (eg Deese 1959 Johnsonet al 1977 Loftus amp Palmer 1974 Loftus amp Pickrell1995) there remains much uncertainty about the neuralmechanisms underlying the creation of false memoriesRecently neuroimaging studies have been conducted tobegin to investigate the neural mechanisms underlyingfalse memories in the production and manifestation atboth the encoding and retrieval levels of the memoryprocess (Cabeza Rao Wagner Mayer amp Schacter 2001Gonsalves amp Paller 2000 Schacter Buckner KoutstaalDale amp Rosen 1997 Schacter et al 1996) It is not sur-prising that in many of the neuroimaging studies theDRM paradigm was adopted for the experimental de-sign since this paradigm robustly elicits false alarms tothematic words This task is not without its limitations asa neuroimaging paradigm however The most pertinentlimitation is that it requires many studied words for onlya few critical lures Despite this limitation modified ver-sions of the DRM paradigm have been used to examinethe neural correlates of true and false recognition

Using both positron emission tomography (PETSchacter et al 1996) and functional magnetic resonanceimaging (f MRI Cabeza et al 2001 Schacter et al1997) researchers have observed that true and falserecognition in the DRM task were associated with simi-lar patterns of activity throughout much of the brainsuggesting that by and large true and false recognitionare associated with similar neural mechanisms Frontal

lobe activity was observed during both true and falserecognition (which differed in both cases from activityfor novel foil items) which suggests that this region maybe involved in strategic monitoring processes that are ac-tivated when participants are attempting to determinewhether a lure word was actually previously presentedRegions such as the anterior cingulate (Cabeza et al2001 Schacter et al 1997) anterior prefrontal cortex(Schacter et al 1997 Schacter et al 1996) left ventro-lateral prefrontal cortex (Cabeza et al 2001) and bilat-eral dorsolateral prefrontal cortex (Cabeza et al 2001Schacter et al 1997) also elicited similar activation pat-terns for true and false recognition

In contrast with these similarities between true andfalse recognition several differences have also been ob-served Cabeza et al (2001) reported that ventromedialprefrontal cortex was associated with more activity forfalse recognition in comparison with true recognitionThis is consistent with numerous neuroimaging studiesthat have suggested that various prefrontal cortical re-gions are differentially involved with various types ofmemory retrieval such as retrieval associated with effortor success or performance of postretrieval verificationor monitoring (eg Buckner Koutstaal Schacter Daleet al 1998 Buckner Koutstaal Schacter Wagner ampRosen 1998 Cabeza amp Nyberg 2000 Rugg FletcherFrith Frackowiak amp Dolan 1996 Schacter et al 1996Wagner Desmond Glover amp Gabrieli 1998) that mightdiffer for true and for false recognition

Also the left parietal cortex appears to respond in agraded fashion depending on the depth of encoding dur-ing study and the amount of information that is recov-ered from memory during test (Nessler Mecklinger ampPenney 2001) Cabeza et al (2001) found that activity inthis region was greatest for retrieval that yielded truememories then for false memories and then for the cor-rect rejection of new items Likewise Henson RuggShallice Josephs amp Dolan (1999) found that this regionshowed a graded response to remember know and newjudgments (see Tulving 1985) Interestingly Wheelerand Buckner (2003) found that the left parietal cortex re-sponds to perceived oldness independent of the numberof times the stimulus was studied or the modality of in-formation retrieved Even when a stimulus was new ifparticipants perceived it as old the left parietal cortexwas active These findings suggest that the left parietalcortex serves as an index of how much contextual infor-mation is retrieved and how much the information is per-ceived as a previously encountered item or event

The medial temporal lobe (MTL) plays a vital role inmemory for facts and events (see Milner Squire amp Kan-del 1998 for a review) As such it is a clear target forneuroimaging research of false memory The only studythat has specifically examined the neural differences inthe MTL between the retrieval of true and false memo-ries (Cabeza et al 2001) again used a version of theDRM paradigm Cabeza et al found an interesting dis-sociation between the parahippocampal gyrus and the

fMRI OF FALSE MEMORY 325

hippocampus The parahippocampal gyrus was associ-ated with greater activity for targets than for either foilsor the critical lures (where activity was similar betweenthe two) In contrast the hippocampus was associatedwith greater activity for targets and critical lures than forfoils Cabeza et al proposed that the parahippocampalgyrus (and not the hippocampus) is sensitive to the sen-sory properties of recovered information and thereforecould potentially differentiate between items seen beforeand items not seen before (even if both are endorsed dur-ing recognition)

Similarly event-related potential (ERP) studies usingthe DRM paradigm showed that unilateral presentationof words during study elicited ipsilateral brain activityduring test for centrally presented studied items but notfor the highly related nonstudied items (Fabiani Stadleramp Wessels 2000) This suggests that true memories ofstudied items leave sensory signals of study experiencesthat make the memory traces distinctive Nonstudieditems lack these sensory signals The P300 component ofthe ERP has also been identified with shorter latenciesfor false recognition of nonstudied lure items in com-parison with true recognition of studied items (MillerBaratta Wynveen amp Rosenfeld 2001)

Using the DRM paradigm Nessler et al (2001) re-ported that the nature of the encoding task determineswhether there are neural differences between true andfalse memories When the encoding task focused on theconceptual similarity of the studied items brain activityfor true and false recognition was similar whereas theydiffered when the encoding task focused on specific itemfeatures For example during the latter task there was alack of frontomedial cortical activity during false recog-nition an area frequently active during feelings of fa-miliarity when there was no such absence of activityduring the first task

Rather than examining the neural mechanisms of falsememories elicited by the DRM paradigm Gonsalves andPaller (2000) investigated false memories using an explicitreality monitoring paradigm Using ERP Gonsalves andPaller showed that during encoding midline occipital andparietal activity during the imagination of a picture wasmore positive for those that were later mistakenly believedto have been actually perceived than for those that werelater correctly identified as having been only imagined Inconsistency with the SMF (see eg Johnson et al 1993)Gonsalves and Paller suggested that the more vivid the vi-sual imagery during encoding the more likely the imag-ined pictures are to become false memories of perceivedpictures During retrieval there was greater midline oc-cipital and parietal activity for accurate perceived picturememories than for false picture memories This suggestedthat the imagery for false memories at retrieval was not asstrong as the imagery for true memories The pattern of re-sults observed in the parietal cortex has also been noted byothers examining true and false memory retrieval and im-agery (Allan Wilding amp Rugg 1998 Cabeza et al 2001Henson et al 1999 Nessler et al 2001)

Thus true and false memories appear to share muchof the same neural activation patterns in similar brain re-gions but evidence also suggests that clear differencesexist Since the nature of the encoding task appears tohave an effect on whether neural differences are elicitedbetween true and false memories it is critical to use par-adigms other than the frequently employed DRM para-digm to further examine these distinguishing markers

One key difference between the DRM and realitymonitoring paradigms is that in the DRM paradigm thefalsely recognized items are not presented at studywhereas in the Gonsalves and Paller (2000) study thefalsely recognized items are presented at least in theform of a cue This may affect activity during retrievaldepending on whether items presented at test correspondin any way to items presented at study

We used the reality monitoring paradigm employed inGonsalves and Pallerrsquos (2000) ERP study to investigatethe neural differences between retrieval processes thatyield true and false memories with f MRI to help providefurther insight into the brain regions and neural circuitryassociated with the creation of false memories

THE PRESENT EXPERIMENT

Using a variation of Gonsalves and Pallerrsquos (2000) ex-perimental paradigm we examined and compared neuralactivity during true and false recognition We report thatactivity in frontal and parietal regions though memoryrelated did not differ between true and false recognitionActivity in several occipital regions and in the anteriorcingulate however differed for true relative to false rec-ollection Occipital activity and anterior cingulate activ-ity differed in numerous respects however which is in-dicative of their different roles in memory retrievalFurthermore we note that although overall there waslittle differential activity in the MTL the activity thatwas observed mirrored the activation pattern observedin the occipital regions

MethodThe experiment was divided into three phases study lie test and

test In the study phase words naming concrete objects were pre-sented auditorily and the participants were asked to vividly imag-ine the object (see Figure 1) An actual picture of the object fol-lowed the presentation of half of the objects

A misinformation task (ie the lie test) was inserted between thisstudy phase and the test phase in an effort to increase the occur-rence of false memories (see eg Loftus et al 1978 Loftus ampPalmer 1974) During the lie test the participants were given a ver-sion of the test phase in which they were encouraged to lie aboutseeing a picture that they had only imagined during the previousstudy phase Here the participants played a game in which the scor-ing was rigged in such a way as to encourage them to claim they hadpreviously seen a picture of an object that they had only imagined

Later during the test phase (in which absolute accuracy wasstressed) having lied and claimed to have seen the picture earlier inthe lie test induced a misattribution error which served as a self-generated source of misinformation As in previous studies of ex-ternal misinformation (eg Loftus et al 1978 Loftus amp Palmer

326 OKADO AND STARK

1974) the participantsrsquo memories appear to have been altered to thepoint that they falsely believed they had actually seen the picturesince the false alarm rate was raised significantly

The test phase was the only in-scanner portion of the experimentIn the test phase the participants heard names of objects they hadseen accompanied by a picture objects that were presented withouta picture and new objects The participants were asked to deter-mine if they had actually seen a picture of the object during thestudy phase We were specifically interested in trials in which theparticipants accurately endorsed perceived pictures as perceived(true memory) and trials in which they inaccurately endorsed imag-ined pictures as perceived (false memory)

The sole purpose of including the lie test was to increase thenumber of false memories available for fMRI analysis which it ap-pears to reliably do In a control experiment conducted outside thescanner half of the studied stimuli were presented in the lie test andhalf were not presented to each participant Overall the false mem-ory rate (ie the rate of inaccurate endorsement of words presentedalone as having been presented along with a picture) during the testphase was significantly higher [t(13) 5 232 p 05] for thoseitems presented in the lie test (17) than for those items that werenot presented in the lie test (13) Although this increases the num-ber of trials available for fMRI data analysis it is nearly impossi-ble to predict what cognitive or neural processes are engaged dur-ing the lie test since true errors made by the participants cannot bedistinguished from deliberate errors Likewise it would be excep-tionally difficult if not impossible to determine in the present par-adigm whether any particular false memory had its source in theoriginal study phase in the lie test or in some combination of bothSince we were interested specifically in examining retrieval pro-cesses yielding false memories and the reconstructive nature of theretrieval process we collected fMRI data only from the test phase

ParticipantsFourteen native English speakers (8 male 6 female) were re-

cruited from the Johns Hopkins University community The partic-ipants were between the ages of 19 and 31 years were right-handedand had no history of neurological or psychiatric illness All theparticipants were naive to the experimental materials and hypothe-ses and gave informed written consent to participate

MaterialsThe stimuli consisted of 450 color photographs of objects on a

white background and auditorily presented names of each object(recorded as sound files by a single female native English speaker)Outside the scanner pictures were presented on a computer screenand the object names were played over headphones Inside thescanner object names were played over pneumatic headphones andresponses were collected using a fiber optic button box

ProcedureThe study phase consisted of a total of 300 trials 150 spoken words

followed by pictures (W1P trials) and 150 spoken words followedby a blank rectangle (W1B trials) The duration of the spoken wordswas approximately 300 msec followed by a delay (~1500 msec)for a total of exactly 1800 msec (see Figure 1) Corresponding pic-tures of the objects or blank rectangles were presented for300 msec followed by a 1500-msec delay The participants wereinstructed to visualize an image of the object and decide if the ob-ject was larger or smaller than a shoebox indicating their responseon the computer keyboard after the word was presented The par-ticipants were informed that a corresponding picture or filler rec-tangle would appear on the screen soon after the word was pre-sented The words presented with pictures and those presented withblank rectangles were counterbalanced In addition to trials withstimuli there were 50 null trials in the form of auditory noise(3600 msec) during which the participant made no response All ofthe stimuli were randomly intermixed The study phase was dividedinto two runs

The lie test consisted of 225 trials (75 of the W1P trials and all150 of the W1B trials from the study phase) of auditorily presentedwords For each word the participants were asked to decide whetheror not they had seen an actual picture of the object in the studyphase On each trial the participants could earn or lose points witheach response with the goal of accumulating as many points as pos-sible (indicated on the screen) Truthful responses were awarded2 6 1 points When the participants lied and endorsed a word stud-ied without a picture they would gain 8 6 1 points on 70 of thetrials and lose 4 6 1 points on 30 of the trials On the latter set oftrials (those checked for accuracy) an animated figure would ap-pear along with a sound effect as the participantrsquos score was re-duced Finally on trials in which the participants lied (or were in-accurate) and did not endorse a trial that had been previouslypresented with a picture they lost 2 6 1 points if the trial waschecked for accuracy but gained 4 6 1 points if it was not checkedTherefore the participants were strongly encouraged to lie and en-dorse words that had not been studied with a picture and they wereonly moderately encouraged to lie and not endorse words that hadbeen studied with a picture The participants were informed thatthey would earn or lose a certain number of points with every re-sponse they made They were given the goal of collecting as manypoints as possible and were told that they would receive feedbackon every trial in the form of a bar graph and the actual number ofpoints collected The participants were informed that although itbenefited them to respond truthfully as to whether or not they hadseen a picture it might benef it them even more to lie if the partic-ular trial was not checked They were told that only a small numberof the trials would be checked The participants indicated their re-sponse on the computer keyboard The trials were self-paced

DOG

300 msec 1500 msec 300 msec 1500 msec

orW+B

W+P

Figure 1 Schematic diagram of the study phase Words (in this example DOG) were pre-sented auditorily followed by a 1500-msec delay Either a picture of the object or a blank rec-tangle was presented for 300 msec followed by a 1500-msec delay The participants imag-ined a picture of the object when the word was presented and a response was required duringthe first 1500-msec delay

fMRI OF FALSE MEMORY 327

The participants were then placed in the fMRI scanner for thetest phase The test phase consisted of 450 trials 150 W1P trials150 W1B trials and 150 trials of spoken words that had not beenpresented at study (foils) Words were presented auditorily every2500 msec The participants were instructed to decide whether ornot they had seen an actual picture of the object in the study phaseThey were instructed that this portion of the experiment pertainedonly to the study phase and to tell the truth to the best of their abil-ity They indicated their ldquoyesrdquo and ldquonordquo responses using a buttonbox In addition there were 75 null trials in the form of auditorynoise (2500 msec) during which the participants made no re-sponse These trials served as a baseline condition for the fMRIdata analysis All of the stimuli were randomly intermixed The testphase was divided into five runs

fMRI MethodsImaging was performed on a Philips Gyroscan 15T MRI scan-

ner equipped with a whole-brain SENSE coil Thirty-f ive T2-weighted triple-oblique functional images were collected per 3-Dvolume using a single-shot echoplanar pulse sequence (64 3 64matrix TE 5 40 msec flip angle 5 90ordm in-plane resolution 5 4 34 mm thickness 5 4 mm TR 5 25 sec) Slices were aligned withthe principal axis of the left and right hippocampus (as determinedby a series of sagittal localizer MRI scans for each participant) Thiswas done to optimize the signal from the medial temporal lobes andto minimize partial-voluming effects so that voxels could be clearlyconstrained to lie within subregions of the MTL A total of 525 vol-umes were collected The task began in synchrony with the acqui-sition of the fifth volume to allow for T1 stabilization After thefunctional scans a high-resolution structural MRI was acquired(MP-RAGE pulse sequence 1 mm3 resolution 150 triple-obliqueaxial slices in the same orientation as the functional images) foranatomical localization

fMRI Data AnalysisImage analysis was performed using Analysis of Functional Neu-

roimages (Cox 1996) Functional MRI data were first resampled intime using a Fourier algorithm to align all slices to a common timebase Functional images were then resampled in space to coregisterthe images and reduce the effects of head motion in three dimen-sions During this process six vectors were created that code for allpossible translations and rotations of the brain fMRI data from alltest runs were concatenated

Following this processing the behavioral data were coded intoeight trial types of interest and a general linear model (GLM) of thefMRI time series data was constructed using these vectors Thestimulus vectors coding for the eight trial types were named ac-cording to study phase condition and participant response as fol-lows (1) SndashS seen pictures endorsed as seen (ldquoyesrdquo response toW1P items) (2) SndashS (no MI) ldquoyesrdquo response to W1P items thatwere not presented during the lie test and therefore received nomisinformation (3) SndashN seen pictures rejected as not seen (ldquonordquoresponse to W1P items) (4) SndashN (no MI) ldquonordquo response to W1Pitems that were not presented during the lie test and therefore re-ceived no misinformation) (5) IndashS imagined pictures endorsed asseen (ldquoyesrdquo response to W1B items) (6) IndashN imagined pictures re-jected as not seen (ldquonordquo response to W1B items) (7) UndashS un-studied words endorsed as seen (ldquoyesrdquo response to foils) and(8) UndashN unstudied words rejected as not seen (ldquonordquo to response tofoils) Unfortunately with only 10 trials per participant on average(range 5 2ndash22) reliable estimates of the hemodynamic response inthe UndashS condition could not be obtained and therefore this trialtype was not included in the group analysis In addition the GLMincluded nuisance vectors coding for first- and second-order driftin the MR signal and for 3-D head motion

The GLM was constructed using a deconvolution technique(Ward 2000) that estimates the impulse response function within

each voxel and performs a multiple linear regression The sum ofthe beta coefficients for the time points corresponding to the ex-pected peak in the hemodynamic response (~25ndash10 sec after stim-ulus onset) was taken as the modelrsquos estimate of the response toeach trial type We should note that this response magnitude is rel-ative to the activity present during the null task (auditory noise) Anactivity level of zero for a particular condition (and a flat hemody-namic response) does not imply a lack of activity in the region dur-ing this condition Rather it implies merely a similar level of ac-tivity in the condition of interest and in the null task which mayvery well be active (Stark amp Squire 2001b)

Initial spatial normalization was done using each participantrsquosstructural MRI to transform data according to the common atlas ofTalairach and Tournoux (1988) This transformation was applied tothe statistical maps of the beta coefficients and in the process thedata were resampled to 25 mm3 The spatially normalized statisti-cal maps of the beta coefficients were blurred using a Gaussian fil-ter with a full-width half maximum of 4 mm to help account forvariations in the functional anatomy across participants

The main analyses involved a voxel-wise two-factor analysis ofvariance (ANOVA) on the beta coefficients with a fixed factor (con-dition all seven trial types) and a random factor (subject) to iden-tify any activation differences between the trial types The ANOVAwas used to define functional regions of interest (ROIs) that demon-strated significant activation differences across any of the seventrial types that passed a voxel-wise threshold of p 01 and a spa-tial extent threshold of 328 mm3 ( p 05 correcting for multiplecomparisons) Post hoc t tests were performed on the average betacoefficients within the ROIs functionally defined by the ANOVAThese t tests were pairwise comparisons of the seven conditionsHemodynamic responses were extracted for three trial types of in-terest (SndashS IndashS and IndashN) for the identif ied regions that showedsignificant differences among these three conditions

The MTL analysis was performed using the ROIndashAL (regions ofinterestndashalignment) technique described by Stark and Okado (2003)The technique begins with a definition of the structures in the MTL(hippocampal region temporopolar perirhinal entorhinal andparahippocampal cortices) bilaterally according to the techniquesdescribed by Insausti et al (1998) The parahippocampal cortexwas further def ined bilaterally as the portion of the parahippo-campal gyrus caudal to the perirhinal cortex and rostral to the sple-nium of the corpus callosum (by this definition the rostral extentincludes only slightly more tissue than as def ined by Pruessneret al 2002) The hippocampal region (the CA fields of the hip-pocampus dentate gyrus and subiculum) was also defined bilater-ally ROIndashAL uses these anatomically defined ROIs to calculate anadditional transformation matrix to fine-tune the cross-participantalignment in the MTL ROIndashAL significantly improves the overlapacross participants increasing statistical power and precision in lo-calization of cross-participant tests In the ROIndashAL analysis avoxel-wise threshold of p 01 and a spatial extent threshold of125 mm3 was used ( p 05 correcting for multiple comparisonswith the MTL alone)

RESULTS

Behavioral ResultsThe overall response rates for each category are shown

in Figure 2a To assess accuracy and false memory ratesoverall discriminability (d cent) scores were calculated fromthe SndashS rate for each of the two types of false alarmrates The dcent for SndashS and IndashS was 138 and the dcent forSndashS and UndashS items was 198 These d cent scores signifi-cantly differed [t(13) 5 534 p 001] as did the over-all rates for the two false alarm conditions [t(13) 5 343

328 OKADO AND STARK

p 01] suggesting that the combination of the realitymonitoring paradigm and the lie test significantly in-creased the false alarm rate to above chance levels andthat the participants were actually experiencing falsememories During the lie test the participants claimed476 of W1B items were studied as pictures Fifty-three percent of the IndashS items from the test phase wereitems that were falsely endorsed during the lie test

Of the seven possible trial types three critical trialtypes were selected a priori for further analyses in an ef-fort to analyze differences between true and false mem-ories SndashS IndashS and IndashN (the left three bars in Figures 2and 3cndashe) Differences between SndashS and IndashS activationsare suggestive of differences between true and falsememory retrieval Both trial types were presented in thestudy phase (SndashS as W1P and IndashS as W1B) and in thelie test and both were endorsed in the test phase (SndashScorrectly endorsed and IndashS incorrectly endorsed) Simi-larly a contrast between IndashS and IndashN can highlight dif-ferences between true and false memory retrieval by in-dicating differences between the false endorsement of animagined picture as being real and the correct rejectionof an imagined picture as having been only imagined (inboth of these conditions the stimuli have been treatedidentically prior to the test phase) Finally by contrast-ing true retrieval of pictures and correct rejection of onlyimagined pictures differences between SndashS and IndashN ac-tivations are suggestive of accurate retrieval However itshould be noted that although this contrast is often usedto assess memory retrieval success it can be contami-nated and weakened by activity associated with inciden-tal encoding during the retrieval task (Stark amp Okado2003)

Figure 2b shows the mean reaction times (RT) for alltrial types In the three conditions of interest the mean

RTs were fastest for SndashS items (1313 msec) and slow-est for IndashS items (1469 msec) A repeated measuresANOVA showed that there was a significant effect ofthese three trial types on RT [F(113) 5 225 p 001]Pairwise t tests showed significant RT differences betweenSndashS and IndashS [156 msec t(13) 5 2474 p 01] be-tween IndashS and IndashN [95 msec t(13) 5 271 p 05] andbetween SndashS and IndashN [61 msec t(13) 5 2306 p 01]

f MRI ResultsA voxel-wise two-factor ANOVA was first conducted

on the whole brain data to identify regions that showedany activation differences among the seven trial typesResults of this analysis identified 12 areas in which ac-tivity significantly differed across trial types (Figure 3a)These areas were treated as functionally defined ROIsand the average hemodynamic response functions foreach trial type were calculated for each region

The 12 regions identified demonstrated three distinctpatterns of activity in the functional ROI analysis withrespect to our three conditions of interest The first no-table pattern observed in the identified regions was sim-ilar increased activity for SndashS and IndashS in comparisonwith IndashN (Figure 3c) This pattern of activity was ob-served in the left parietal cortex and left frontal regionsRegions in which SndashS and IndashN differed significantly andIndashS and IndashN differed significantly included the left pari-etal cortex [BA 73940 t(13) 5 329 p 01 andt(13) 5 257 p 05 respectively] left precentral gyrus[BA 9 t(13) 5 245 p 05 and t(13) 5 319 p 01respectively] and left inferior frontal gyrus [BA 1046t(13) 5 261 p 05 and t(13) 5 298 p 05 respec-tively] The left caudate showed this similar pattern ofactivation with SndashS and IndashS demonstrating increasedactivity in comparison with IndashN but showed no signifi-

Figure 2 Behavioral data from the test phase Percentage of trials endorsed as having beenpresented with a picture at time of study (a) and reaction time for the decision (b) are plot-ted for each trial type as means across all 14 participants Error bars indicate the standarderrors of the means

fMRI OF FALSE MEMORY 329

Figure 3 (a and b) Locations and sample hemodynamic responses for regions where activity at testvaried among the seven trial types analyzed (there were too few trials in the UndashS condition to ana-lyze) (a) The 12 regions where activity varied as a function of trial type are shown as colored overlayson 3-D renderings of a brain A left parietal cortex (BA 73940) B left precentral gyrus (BA 9) Cleft inferior frontal gyrus (BA 1046) D left caudate E right middle occipital gyrus (BA 1819) Fleft cuneus (BA 171819) G left lingual gyrus (BA 19) H left fusiform gyrus (1819) I bilaterallingual gyrus (BA 18) J bilateral cuneus (BA 17182730) K right posterior parahippocampalgyrus (BA 30353637) L right anterior cingulate gyrus (BA 6932) (b) Average hemodynamic re-sponse functions (sum of beta coefficients vs image acquisition [TR]) for the three trial types of in-terest (SndashS IndashS and IndashN) are shown for three sample regions (left precentral gyrus left cuneus andright anterior cingulate) that demonstrate the patterns of activity observed (cndashe) Activity for all seventrial types analyzed in all 12 functionally defined ROIs Bars show the mean fMRI response (sum ofbeta coefficients) across participants and error bars show the standard errors of the means Aster-isks indicate significant differences in activity between conditions of interest (SndashS IndashS and IndashN) (c)Regions demonstrating the first pattern of activity included A left parietal cortex B left precentralgyrus C left inferior frontal gyrus and D left caudate (d) Regions demonstrating the second pat-tern of activity included E right middle occipital gyrus F left cuneus G left lingual gyrus H leftfusiform gyrus I bilateral lingual gyrus J bilateral cuneus and K right posterior parahippocampalgyrus (e) The third pattern of activity was observed in L right anterior cingulate gyrus

B L precentralgyrus (BA 9)

F L cuneus(BA 171819)

L R anteriorcingulate (BA 6932)

TR (25 sec)

beta

wei

ghts

b

c

d

050

ndash05ndash1

ndash15

08

04

0

05

0

ndash05

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

4

2

0

2

2

0

ndash2

ndash4

2

0

ndash2

ndash4

4

2

0

ndash2

A B C D

E F G H

I J K e L

S-S

I-S

I-N

S-S(No MI)

S-N

S-N(No MI)

U-N

1 2 3 4 5 6 7 8

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

324 OKADO AND STARK

contextual detail (Johnson 1997 Johnson amp Raye 1981)Reality monitoring paradigms are used to investigatehow people discriminate externally generated and inter-nally generated memories Behavioral studies show thatfailures in reality monitoring can occur when events thatwere imagined are remembered with considerable sen-sory detail and mistakenly believed to have actually oc-curred (eg Johnson amp Raye 1981 Johnson Raye Wangamp Taylor 1979 Johnson Taylor amp Raye 1977) Be-cause the brain mechanisms involved in mental imagerymay be similar to those used in perceiving (see egFarah 1989 Kosslyn et al 1999 Wheeler Petersen ampBuckner 2000) it is conceivable that such an overlapcan cause imagined events to be misperceived as real

The propensity of misattribution can be increased bysuggestion Loftus and colleagues discovered that whenpeople are given suggestive and misleading informationabout a previous event their recollections of the originalevent can be altered by the misinformation (eg Loftuset al 1978 Loftus amp Palmer 1974) It has been sug-gested that the misinformation may be replacing theoriginal memory (Loftus et al 1978) or it may be pro-viding information that competes with the original mem-ory during retrieval (McCloskey amp Zaragoza 1985) Re-gardless it is clear that misinformation decreases thelikelihood of accurate memory retrieval and increasesthe chances that misinformation will be misattributed tothe original event

Although behavioral studies have extensively investi-gated and provided evidence for the existence and cre-ation of false memories (eg Deese 1959 Johnsonet al 1977 Loftus amp Palmer 1974 Loftus amp Pickrell1995) there remains much uncertainty about the neuralmechanisms underlying the creation of false memoriesRecently neuroimaging studies have been conducted tobegin to investigate the neural mechanisms underlyingfalse memories in the production and manifestation atboth the encoding and retrieval levels of the memoryprocess (Cabeza Rao Wagner Mayer amp Schacter 2001Gonsalves amp Paller 2000 Schacter Buckner KoutstaalDale amp Rosen 1997 Schacter et al 1996) It is not sur-prising that in many of the neuroimaging studies theDRM paradigm was adopted for the experimental de-sign since this paradigm robustly elicits false alarms tothematic words This task is not without its limitations asa neuroimaging paradigm however The most pertinentlimitation is that it requires many studied words for onlya few critical lures Despite this limitation modified ver-sions of the DRM paradigm have been used to examinethe neural correlates of true and false recognition

Using both positron emission tomography (PETSchacter et al 1996) and functional magnetic resonanceimaging (f MRI Cabeza et al 2001 Schacter et al1997) researchers have observed that true and falserecognition in the DRM task were associated with simi-lar patterns of activity throughout much of the brainsuggesting that by and large true and false recognitionare associated with similar neural mechanisms Frontal

lobe activity was observed during both true and falserecognition (which differed in both cases from activityfor novel foil items) which suggests that this region maybe involved in strategic monitoring processes that are ac-tivated when participants are attempting to determinewhether a lure word was actually previously presentedRegions such as the anterior cingulate (Cabeza et al2001 Schacter et al 1997) anterior prefrontal cortex(Schacter et al 1997 Schacter et al 1996) left ventro-lateral prefrontal cortex (Cabeza et al 2001) and bilat-eral dorsolateral prefrontal cortex (Cabeza et al 2001Schacter et al 1997) also elicited similar activation pat-terns for true and false recognition

In contrast with these similarities between true andfalse recognition several differences have also been ob-served Cabeza et al (2001) reported that ventromedialprefrontal cortex was associated with more activity forfalse recognition in comparison with true recognitionThis is consistent with numerous neuroimaging studiesthat have suggested that various prefrontal cortical re-gions are differentially involved with various types ofmemory retrieval such as retrieval associated with effortor success or performance of postretrieval verificationor monitoring (eg Buckner Koutstaal Schacter Daleet al 1998 Buckner Koutstaal Schacter Wagner ampRosen 1998 Cabeza amp Nyberg 2000 Rugg FletcherFrith Frackowiak amp Dolan 1996 Schacter et al 1996Wagner Desmond Glover amp Gabrieli 1998) that mightdiffer for true and for false recognition

Also the left parietal cortex appears to respond in agraded fashion depending on the depth of encoding dur-ing study and the amount of information that is recov-ered from memory during test (Nessler Mecklinger ampPenney 2001) Cabeza et al (2001) found that activity inthis region was greatest for retrieval that yielded truememories then for false memories and then for the cor-rect rejection of new items Likewise Henson RuggShallice Josephs amp Dolan (1999) found that this regionshowed a graded response to remember know and newjudgments (see Tulving 1985) Interestingly Wheelerand Buckner (2003) found that the left parietal cortex re-sponds to perceived oldness independent of the numberof times the stimulus was studied or the modality of in-formation retrieved Even when a stimulus was new ifparticipants perceived it as old the left parietal cortexwas active These findings suggest that the left parietalcortex serves as an index of how much contextual infor-mation is retrieved and how much the information is per-ceived as a previously encountered item or event

The medial temporal lobe (MTL) plays a vital role inmemory for facts and events (see Milner Squire amp Kan-del 1998 for a review) As such it is a clear target forneuroimaging research of false memory The only studythat has specifically examined the neural differences inthe MTL between the retrieval of true and false memo-ries (Cabeza et al 2001) again used a version of theDRM paradigm Cabeza et al found an interesting dis-sociation between the parahippocampal gyrus and the

fMRI OF FALSE MEMORY 325

hippocampus The parahippocampal gyrus was associ-ated with greater activity for targets than for either foilsor the critical lures (where activity was similar betweenthe two) In contrast the hippocampus was associatedwith greater activity for targets and critical lures than forfoils Cabeza et al proposed that the parahippocampalgyrus (and not the hippocampus) is sensitive to the sen-sory properties of recovered information and thereforecould potentially differentiate between items seen beforeand items not seen before (even if both are endorsed dur-ing recognition)

Similarly event-related potential (ERP) studies usingthe DRM paradigm showed that unilateral presentationof words during study elicited ipsilateral brain activityduring test for centrally presented studied items but notfor the highly related nonstudied items (Fabiani Stadleramp Wessels 2000) This suggests that true memories ofstudied items leave sensory signals of study experiencesthat make the memory traces distinctive Nonstudieditems lack these sensory signals The P300 component ofthe ERP has also been identified with shorter latenciesfor false recognition of nonstudied lure items in com-parison with true recognition of studied items (MillerBaratta Wynveen amp Rosenfeld 2001)

Using the DRM paradigm Nessler et al (2001) re-ported that the nature of the encoding task determineswhether there are neural differences between true andfalse memories When the encoding task focused on theconceptual similarity of the studied items brain activityfor true and false recognition was similar whereas theydiffered when the encoding task focused on specific itemfeatures For example during the latter task there was alack of frontomedial cortical activity during false recog-nition an area frequently active during feelings of fa-miliarity when there was no such absence of activityduring the first task

Rather than examining the neural mechanisms of falsememories elicited by the DRM paradigm Gonsalves andPaller (2000) investigated false memories using an explicitreality monitoring paradigm Using ERP Gonsalves andPaller showed that during encoding midline occipital andparietal activity during the imagination of a picture wasmore positive for those that were later mistakenly believedto have been actually perceived than for those that werelater correctly identified as having been only imagined Inconsistency with the SMF (see eg Johnson et al 1993)Gonsalves and Paller suggested that the more vivid the vi-sual imagery during encoding the more likely the imag-ined pictures are to become false memories of perceivedpictures During retrieval there was greater midline oc-cipital and parietal activity for accurate perceived picturememories than for false picture memories This suggestedthat the imagery for false memories at retrieval was not asstrong as the imagery for true memories The pattern of re-sults observed in the parietal cortex has also been noted byothers examining true and false memory retrieval and im-agery (Allan Wilding amp Rugg 1998 Cabeza et al 2001Henson et al 1999 Nessler et al 2001)

Thus true and false memories appear to share muchof the same neural activation patterns in similar brain re-gions but evidence also suggests that clear differencesexist Since the nature of the encoding task appears tohave an effect on whether neural differences are elicitedbetween true and false memories it is critical to use par-adigms other than the frequently employed DRM para-digm to further examine these distinguishing markers

One key difference between the DRM and realitymonitoring paradigms is that in the DRM paradigm thefalsely recognized items are not presented at studywhereas in the Gonsalves and Paller (2000) study thefalsely recognized items are presented at least in theform of a cue This may affect activity during retrievaldepending on whether items presented at test correspondin any way to items presented at study

We used the reality monitoring paradigm employed inGonsalves and Pallerrsquos (2000) ERP study to investigatethe neural differences between retrieval processes thatyield true and false memories with f MRI to help providefurther insight into the brain regions and neural circuitryassociated with the creation of false memories

THE PRESENT EXPERIMENT

Using a variation of Gonsalves and Pallerrsquos (2000) ex-perimental paradigm we examined and compared neuralactivity during true and false recognition We report thatactivity in frontal and parietal regions though memoryrelated did not differ between true and false recognitionActivity in several occipital regions and in the anteriorcingulate however differed for true relative to false rec-ollection Occipital activity and anterior cingulate activ-ity differed in numerous respects however which is in-dicative of their different roles in memory retrievalFurthermore we note that although overall there waslittle differential activity in the MTL the activity thatwas observed mirrored the activation pattern observedin the occipital regions

MethodThe experiment was divided into three phases study lie test and

test In the study phase words naming concrete objects were pre-sented auditorily and the participants were asked to vividly imag-ine the object (see Figure 1) An actual picture of the object fol-lowed the presentation of half of the objects

A misinformation task (ie the lie test) was inserted between thisstudy phase and the test phase in an effort to increase the occur-rence of false memories (see eg Loftus et al 1978 Loftus ampPalmer 1974) During the lie test the participants were given a ver-sion of the test phase in which they were encouraged to lie aboutseeing a picture that they had only imagined during the previousstudy phase Here the participants played a game in which the scor-ing was rigged in such a way as to encourage them to claim they hadpreviously seen a picture of an object that they had only imagined

Later during the test phase (in which absolute accuracy wasstressed) having lied and claimed to have seen the picture earlier inthe lie test induced a misattribution error which served as a self-generated source of misinformation As in previous studies of ex-ternal misinformation (eg Loftus et al 1978 Loftus amp Palmer

326 OKADO AND STARK

1974) the participantsrsquo memories appear to have been altered to thepoint that they falsely believed they had actually seen the picturesince the false alarm rate was raised significantly

The test phase was the only in-scanner portion of the experimentIn the test phase the participants heard names of objects they hadseen accompanied by a picture objects that were presented withouta picture and new objects The participants were asked to deter-mine if they had actually seen a picture of the object during thestudy phase We were specifically interested in trials in which theparticipants accurately endorsed perceived pictures as perceived(true memory) and trials in which they inaccurately endorsed imag-ined pictures as perceived (false memory)

The sole purpose of including the lie test was to increase thenumber of false memories available for fMRI analysis which it ap-pears to reliably do In a control experiment conducted outside thescanner half of the studied stimuli were presented in the lie test andhalf were not presented to each participant Overall the false mem-ory rate (ie the rate of inaccurate endorsement of words presentedalone as having been presented along with a picture) during the testphase was significantly higher [t(13) 5 232 p 05] for thoseitems presented in the lie test (17) than for those items that werenot presented in the lie test (13) Although this increases the num-ber of trials available for fMRI data analysis it is nearly impossi-ble to predict what cognitive or neural processes are engaged dur-ing the lie test since true errors made by the participants cannot bedistinguished from deliberate errors Likewise it would be excep-tionally difficult if not impossible to determine in the present par-adigm whether any particular false memory had its source in theoriginal study phase in the lie test or in some combination of bothSince we were interested specifically in examining retrieval pro-cesses yielding false memories and the reconstructive nature of theretrieval process we collected fMRI data only from the test phase

ParticipantsFourteen native English speakers (8 male 6 female) were re-

cruited from the Johns Hopkins University community The partic-ipants were between the ages of 19 and 31 years were right-handedand had no history of neurological or psychiatric illness All theparticipants were naive to the experimental materials and hypothe-ses and gave informed written consent to participate

MaterialsThe stimuli consisted of 450 color photographs of objects on a

white background and auditorily presented names of each object(recorded as sound files by a single female native English speaker)Outside the scanner pictures were presented on a computer screenand the object names were played over headphones Inside thescanner object names were played over pneumatic headphones andresponses were collected using a fiber optic button box

ProcedureThe study phase consisted of a total of 300 trials 150 spoken words

followed by pictures (W1P trials) and 150 spoken words followedby a blank rectangle (W1B trials) The duration of the spoken wordswas approximately 300 msec followed by a delay (~1500 msec)for a total of exactly 1800 msec (see Figure 1) Corresponding pic-tures of the objects or blank rectangles were presented for300 msec followed by a 1500-msec delay The participants wereinstructed to visualize an image of the object and decide if the ob-ject was larger or smaller than a shoebox indicating their responseon the computer keyboard after the word was presented The par-ticipants were informed that a corresponding picture or filler rec-tangle would appear on the screen soon after the word was pre-sented The words presented with pictures and those presented withblank rectangles were counterbalanced In addition to trials withstimuli there were 50 null trials in the form of auditory noise(3600 msec) during which the participant made no response All ofthe stimuli were randomly intermixed The study phase was dividedinto two runs

The lie test consisted of 225 trials (75 of the W1P trials and all150 of the W1B trials from the study phase) of auditorily presentedwords For each word the participants were asked to decide whetheror not they had seen an actual picture of the object in the studyphase On each trial the participants could earn or lose points witheach response with the goal of accumulating as many points as pos-sible (indicated on the screen) Truthful responses were awarded2 6 1 points When the participants lied and endorsed a word stud-ied without a picture they would gain 8 6 1 points on 70 of thetrials and lose 4 6 1 points on 30 of the trials On the latter set oftrials (those checked for accuracy) an animated figure would ap-pear along with a sound effect as the participantrsquos score was re-duced Finally on trials in which the participants lied (or were in-accurate) and did not endorse a trial that had been previouslypresented with a picture they lost 2 6 1 points if the trial waschecked for accuracy but gained 4 6 1 points if it was not checkedTherefore the participants were strongly encouraged to lie and en-dorse words that had not been studied with a picture and they wereonly moderately encouraged to lie and not endorse words that hadbeen studied with a picture The participants were informed thatthey would earn or lose a certain number of points with every re-sponse they made They were given the goal of collecting as manypoints as possible and were told that they would receive feedbackon every trial in the form of a bar graph and the actual number ofpoints collected The participants were informed that although itbenefited them to respond truthfully as to whether or not they hadseen a picture it might benef it them even more to lie if the partic-ular trial was not checked They were told that only a small numberof the trials would be checked The participants indicated their re-sponse on the computer keyboard The trials were self-paced

DOG

300 msec 1500 msec 300 msec 1500 msec

orW+B

W+P

Figure 1 Schematic diagram of the study phase Words (in this example DOG) were pre-sented auditorily followed by a 1500-msec delay Either a picture of the object or a blank rec-tangle was presented for 300 msec followed by a 1500-msec delay The participants imag-ined a picture of the object when the word was presented and a response was required duringthe first 1500-msec delay

fMRI OF FALSE MEMORY 327

The participants were then placed in the fMRI scanner for thetest phase The test phase consisted of 450 trials 150 W1P trials150 W1B trials and 150 trials of spoken words that had not beenpresented at study (foils) Words were presented auditorily every2500 msec The participants were instructed to decide whether ornot they had seen an actual picture of the object in the study phaseThey were instructed that this portion of the experiment pertainedonly to the study phase and to tell the truth to the best of their abil-ity They indicated their ldquoyesrdquo and ldquonordquo responses using a buttonbox In addition there were 75 null trials in the form of auditorynoise (2500 msec) during which the participants made no re-sponse These trials served as a baseline condition for the fMRIdata analysis All of the stimuli were randomly intermixed The testphase was divided into five runs

fMRI MethodsImaging was performed on a Philips Gyroscan 15T MRI scan-

ner equipped with a whole-brain SENSE coil Thirty-f ive T2-weighted triple-oblique functional images were collected per 3-Dvolume using a single-shot echoplanar pulse sequence (64 3 64matrix TE 5 40 msec flip angle 5 90ordm in-plane resolution 5 4 34 mm thickness 5 4 mm TR 5 25 sec) Slices were aligned withthe principal axis of the left and right hippocampus (as determinedby a series of sagittal localizer MRI scans for each participant) Thiswas done to optimize the signal from the medial temporal lobes andto minimize partial-voluming effects so that voxels could be clearlyconstrained to lie within subregions of the MTL A total of 525 vol-umes were collected The task began in synchrony with the acqui-sition of the fifth volume to allow for T1 stabilization After thefunctional scans a high-resolution structural MRI was acquired(MP-RAGE pulse sequence 1 mm3 resolution 150 triple-obliqueaxial slices in the same orientation as the functional images) foranatomical localization

fMRI Data AnalysisImage analysis was performed using Analysis of Functional Neu-

roimages (Cox 1996) Functional MRI data were first resampled intime using a Fourier algorithm to align all slices to a common timebase Functional images were then resampled in space to coregisterthe images and reduce the effects of head motion in three dimen-sions During this process six vectors were created that code for allpossible translations and rotations of the brain fMRI data from alltest runs were concatenated

Following this processing the behavioral data were coded intoeight trial types of interest and a general linear model (GLM) of thefMRI time series data was constructed using these vectors Thestimulus vectors coding for the eight trial types were named ac-cording to study phase condition and participant response as fol-lows (1) SndashS seen pictures endorsed as seen (ldquoyesrdquo response toW1P items) (2) SndashS (no MI) ldquoyesrdquo response to W1P items thatwere not presented during the lie test and therefore received nomisinformation (3) SndashN seen pictures rejected as not seen (ldquonordquoresponse to W1P items) (4) SndashN (no MI) ldquonordquo response to W1Pitems that were not presented during the lie test and therefore re-ceived no misinformation) (5) IndashS imagined pictures endorsed asseen (ldquoyesrdquo response to W1B items) (6) IndashN imagined pictures re-jected as not seen (ldquonordquo response to W1B items) (7) UndashS un-studied words endorsed as seen (ldquoyesrdquo response to foils) and(8) UndashN unstudied words rejected as not seen (ldquonordquo to response tofoils) Unfortunately with only 10 trials per participant on average(range 5 2ndash22) reliable estimates of the hemodynamic response inthe UndashS condition could not be obtained and therefore this trialtype was not included in the group analysis In addition the GLMincluded nuisance vectors coding for first- and second-order driftin the MR signal and for 3-D head motion

The GLM was constructed using a deconvolution technique(Ward 2000) that estimates the impulse response function within

each voxel and performs a multiple linear regression The sum ofthe beta coefficients for the time points corresponding to the ex-pected peak in the hemodynamic response (~25ndash10 sec after stim-ulus onset) was taken as the modelrsquos estimate of the response toeach trial type We should note that this response magnitude is rel-ative to the activity present during the null task (auditory noise) Anactivity level of zero for a particular condition (and a flat hemody-namic response) does not imply a lack of activity in the region dur-ing this condition Rather it implies merely a similar level of ac-tivity in the condition of interest and in the null task which mayvery well be active (Stark amp Squire 2001b)

Initial spatial normalization was done using each participantrsquosstructural MRI to transform data according to the common atlas ofTalairach and Tournoux (1988) This transformation was applied tothe statistical maps of the beta coefficients and in the process thedata were resampled to 25 mm3 The spatially normalized statisti-cal maps of the beta coefficients were blurred using a Gaussian fil-ter with a full-width half maximum of 4 mm to help account forvariations in the functional anatomy across participants

The main analyses involved a voxel-wise two-factor analysis ofvariance (ANOVA) on the beta coefficients with a fixed factor (con-dition all seven trial types) and a random factor (subject) to iden-tify any activation differences between the trial types The ANOVAwas used to define functional regions of interest (ROIs) that demon-strated significant activation differences across any of the seventrial types that passed a voxel-wise threshold of p 01 and a spa-tial extent threshold of 328 mm3 ( p 05 correcting for multiplecomparisons) Post hoc t tests were performed on the average betacoefficients within the ROIs functionally defined by the ANOVAThese t tests were pairwise comparisons of the seven conditionsHemodynamic responses were extracted for three trial types of in-terest (SndashS IndashS and IndashN) for the identif ied regions that showedsignificant differences among these three conditions

The MTL analysis was performed using the ROIndashAL (regions ofinterestndashalignment) technique described by Stark and Okado (2003)The technique begins with a definition of the structures in the MTL(hippocampal region temporopolar perirhinal entorhinal andparahippocampal cortices) bilaterally according to the techniquesdescribed by Insausti et al (1998) The parahippocampal cortexwas further def ined bilaterally as the portion of the parahippo-campal gyrus caudal to the perirhinal cortex and rostral to the sple-nium of the corpus callosum (by this definition the rostral extentincludes only slightly more tissue than as def ined by Pruessneret al 2002) The hippocampal region (the CA fields of the hip-pocampus dentate gyrus and subiculum) was also defined bilater-ally ROIndashAL uses these anatomically defined ROIs to calculate anadditional transformation matrix to fine-tune the cross-participantalignment in the MTL ROIndashAL significantly improves the overlapacross participants increasing statistical power and precision in lo-calization of cross-participant tests In the ROIndashAL analysis avoxel-wise threshold of p 01 and a spatial extent threshold of125 mm3 was used ( p 05 correcting for multiple comparisonswith the MTL alone)

RESULTS

Behavioral ResultsThe overall response rates for each category are shown

in Figure 2a To assess accuracy and false memory ratesoverall discriminability (d cent) scores were calculated fromthe SndashS rate for each of the two types of false alarmrates The dcent for SndashS and IndashS was 138 and the dcent forSndashS and UndashS items was 198 These d cent scores signifi-cantly differed [t(13) 5 534 p 001] as did the over-all rates for the two false alarm conditions [t(13) 5 343

328 OKADO AND STARK

p 01] suggesting that the combination of the realitymonitoring paradigm and the lie test significantly in-creased the false alarm rate to above chance levels andthat the participants were actually experiencing falsememories During the lie test the participants claimed476 of W1B items were studied as pictures Fifty-three percent of the IndashS items from the test phase wereitems that were falsely endorsed during the lie test

Of the seven possible trial types three critical trialtypes were selected a priori for further analyses in an ef-fort to analyze differences between true and false mem-ories SndashS IndashS and IndashN (the left three bars in Figures 2and 3cndashe) Differences between SndashS and IndashS activationsare suggestive of differences between true and falsememory retrieval Both trial types were presented in thestudy phase (SndashS as W1P and IndashS as W1B) and in thelie test and both were endorsed in the test phase (SndashScorrectly endorsed and IndashS incorrectly endorsed) Simi-larly a contrast between IndashS and IndashN can highlight dif-ferences between true and false memory retrieval by in-dicating differences between the false endorsement of animagined picture as being real and the correct rejectionof an imagined picture as having been only imagined (inboth of these conditions the stimuli have been treatedidentically prior to the test phase) Finally by contrast-ing true retrieval of pictures and correct rejection of onlyimagined pictures differences between SndashS and IndashN ac-tivations are suggestive of accurate retrieval However itshould be noted that although this contrast is often usedto assess memory retrieval success it can be contami-nated and weakened by activity associated with inciden-tal encoding during the retrieval task (Stark amp Okado2003)

Figure 2b shows the mean reaction times (RT) for alltrial types In the three conditions of interest the mean

RTs were fastest for SndashS items (1313 msec) and slow-est for IndashS items (1469 msec) A repeated measuresANOVA showed that there was a significant effect ofthese three trial types on RT [F(113) 5 225 p 001]Pairwise t tests showed significant RT differences betweenSndashS and IndashS [156 msec t(13) 5 2474 p 01] be-tween IndashS and IndashN [95 msec t(13) 5 271 p 05] andbetween SndashS and IndashN [61 msec t(13) 5 2306 p 01]

f MRI ResultsA voxel-wise two-factor ANOVA was first conducted

on the whole brain data to identify regions that showedany activation differences among the seven trial typesResults of this analysis identified 12 areas in which ac-tivity significantly differed across trial types (Figure 3a)These areas were treated as functionally defined ROIsand the average hemodynamic response functions foreach trial type were calculated for each region

The 12 regions identified demonstrated three distinctpatterns of activity in the functional ROI analysis withrespect to our three conditions of interest The first no-table pattern observed in the identified regions was sim-ilar increased activity for SndashS and IndashS in comparisonwith IndashN (Figure 3c) This pattern of activity was ob-served in the left parietal cortex and left frontal regionsRegions in which SndashS and IndashN differed significantly andIndashS and IndashN differed significantly included the left pari-etal cortex [BA 73940 t(13) 5 329 p 01 andt(13) 5 257 p 05 respectively] left precentral gyrus[BA 9 t(13) 5 245 p 05 and t(13) 5 319 p 01respectively] and left inferior frontal gyrus [BA 1046t(13) 5 261 p 05 and t(13) 5 298 p 05 respec-tively] The left caudate showed this similar pattern ofactivation with SndashS and IndashS demonstrating increasedactivity in comparison with IndashN but showed no signifi-

Figure 2 Behavioral data from the test phase Percentage of trials endorsed as having beenpresented with a picture at time of study (a) and reaction time for the decision (b) are plot-ted for each trial type as means across all 14 participants Error bars indicate the standarderrors of the means

fMRI OF FALSE MEMORY 329

Figure 3 (a and b) Locations and sample hemodynamic responses for regions where activity at testvaried among the seven trial types analyzed (there were too few trials in the UndashS condition to ana-lyze) (a) The 12 regions where activity varied as a function of trial type are shown as colored overlayson 3-D renderings of a brain A left parietal cortex (BA 73940) B left precentral gyrus (BA 9) Cleft inferior frontal gyrus (BA 1046) D left caudate E right middle occipital gyrus (BA 1819) Fleft cuneus (BA 171819) G left lingual gyrus (BA 19) H left fusiform gyrus (1819) I bilaterallingual gyrus (BA 18) J bilateral cuneus (BA 17182730) K right posterior parahippocampalgyrus (BA 30353637) L right anterior cingulate gyrus (BA 6932) (b) Average hemodynamic re-sponse functions (sum of beta coefficients vs image acquisition [TR]) for the three trial types of in-terest (SndashS IndashS and IndashN) are shown for three sample regions (left precentral gyrus left cuneus andright anterior cingulate) that demonstrate the patterns of activity observed (cndashe) Activity for all seventrial types analyzed in all 12 functionally defined ROIs Bars show the mean fMRI response (sum ofbeta coefficients) across participants and error bars show the standard errors of the means Aster-isks indicate significant differences in activity between conditions of interest (SndashS IndashS and IndashN) (c)Regions demonstrating the first pattern of activity included A left parietal cortex B left precentralgyrus C left inferior frontal gyrus and D left caudate (d) Regions demonstrating the second pat-tern of activity included E right middle occipital gyrus F left cuneus G left lingual gyrus H leftfusiform gyrus I bilateral lingual gyrus J bilateral cuneus and K right posterior parahippocampalgyrus (e) The third pattern of activity was observed in L right anterior cingulate gyrus

B L precentralgyrus (BA 9)

F L cuneus(BA 171819)

L R anteriorcingulate (BA 6932)

TR (25 sec)

beta

wei

ghts

b

c

d

050

ndash05ndash1

ndash15

08

04

0

05

0

ndash05

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

4

2

0

2

2

0

ndash2

ndash4

2

0

ndash2

ndash4

4

2

0

ndash2

A B C D

E F G H

I J K e L

S-S

I-S

I-N

S-S(No MI)

S-N

S-N(No MI)

U-N

1 2 3 4 5 6 7 8

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

fMRI OF FALSE MEMORY 325

hippocampus The parahippocampal gyrus was associ-ated with greater activity for targets than for either foilsor the critical lures (where activity was similar betweenthe two) In contrast the hippocampus was associatedwith greater activity for targets and critical lures than forfoils Cabeza et al proposed that the parahippocampalgyrus (and not the hippocampus) is sensitive to the sen-sory properties of recovered information and thereforecould potentially differentiate between items seen beforeand items not seen before (even if both are endorsed dur-ing recognition)

Similarly event-related potential (ERP) studies usingthe DRM paradigm showed that unilateral presentationof words during study elicited ipsilateral brain activityduring test for centrally presented studied items but notfor the highly related nonstudied items (Fabiani Stadleramp Wessels 2000) This suggests that true memories ofstudied items leave sensory signals of study experiencesthat make the memory traces distinctive Nonstudieditems lack these sensory signals The P300 component ofthe ERP has also been identified with shorter latenciesfor false recognition of nonstudied lure items in com-parison with true recognition of studied items (MillerBaratta Wynveen amp Rosenfeld 2001)

Using the DRM paradigm Nessler et al (2001) re-ported that the nature of the encoding task determineswhether there are neural differences between true andfalse memories When the encoding task focused on theconceptual similarity of the studied items brain activityfor true and false recognition was similar whereas theydiffered when the encoding task focused on specific itemfeatures For example during the latter task there was alack of frontomedial cortical activity during false recog-nition an area frequently active during feelings of fa-miliarity when there was no such absence of activityduring the first task

Rather than examining the neural mechanisms of falsememories elicited by the DRM paradigm Gonsalves andPaller (2000) investigated false memories using an explicitreality monitoring paradigm Using ERP Gonsalves andPaller showed that during encoding midline occipital andparietal activity during the imagination of a picture wasmore positive for those that were later mistakenly believedto have been actually perceived than for those that werelater correctly identified as having been only imagined Inconsistency with the SMF (see eg Johnson et al 1993)Gonsalves and Paller suggested that the more vivid the vi-sual imagery during encoding the more likely the imag-ined pictures are to become false memories of perceivedpictures During retrieval there was greater midline oc-cipital and parietal activity for accurate perceived picturememories than for false picture memories This suggestedthat the imagery for false memories at retrieval was not asstrong as the imagery for true memories The pattern of re-sults observed in the parietal cortex has also been noted byothers examining true and false memory retrieval and im-agery (Allan Wilding amp Rugg 1998 Cabeza et al 2001Henson et al 1999 Nessler et al 2001)

Thus true and false memories appear to share muchof the same neural activation patterns in similar brain re-gions but evidence also suggests that clear differencesexist Since the nature of the encoding task appears tohave an effect on whether neural differences are elicitedbetween true and false memories it is critical to use par-adigms other than the frequently employed DRM para-digm to further examine these distinguishing markers

One key difference between the DRM and realitymonitoring paradigms is that in the DRM paradigm thefalsely recognized items are not presented at studywhereas in the Gonsalves and Paller (2000) study thefalsely recognized items are presented at least in theform of a cue This may affect activity during retrievaldepending on whether items presented at test correspondin any way to items presented at study

We used the reality monitoring paradigm employed inGonsalves and Pallerrsquos (2000) ERP study to investigatethe neural differences between retrieval processes thatyield true and false memories with f MRI to help providefurther insight into the brain regions and neural circuitryassociated with the creation of false memories

THE PRESENT EXPERIMENT

Using a variation of Gonsalves and Pallerrsquos (2000) ex-perimental paradigm we examined and compared neuralactivity during true and false recognition We report thatactivity in frontal and parietal regions though memoryrelated did not differ between true and false recognitionActivity in several occipital regions and in the anteriorcingulate however differed for true relative to false rec-ollection Occipital activity and anterior cingulate activ-ity differed in numerous respects however which is in-dicative of their different roles in memory retrievalFurthermore we note that although overall there waslittle differential activity in the MTL the activity thatwas observed mirrored the activation pattern observedin the occipital regions

MethodThe experiment was divided into three phases study lie test and

test In the study phase words naming concrete objects were pre-sented auditorily and the participants were asked to vividly imag-ine the object (see Figure 1) An actual picture of the object fol-lowed the presentation of half of the objects

A misinformation task (ie the lie test) was inserted between thisstudy phase and the test phase in an effort to increase the occur-rence of false memories (see eg Loftus et al 1978 Loftus ampPalmer 1974) During the lie test the participants were given a ver-sion of the test phase in which they were encouraged to lie aboutseeing a picture that they had only imagined during the previousstudy phase Here the participants played a game in which the scor-ing was rigged in such a way as to encourage them to claim they hadpreviously seen a picture of an object that they had only imagined

Later during the test phase (in which absolute accuracy wasstressed) having lied and claimed to have seen the picture earlier inthe lie test induced a misattribution error which served as a self-generated source of misinformation As in previous studies of ex-ternal misinformation (eg Loftus et al 1978 Loftus amp Palmer

326 OKADO AND STARK

1974) the participantsrsquo memories appear to have been altered to thepoint that they falsely believed they had actually seen the picturesince the false alarm rate was raised significantly

The test phase was the only in-scanner portion of the experimentIn the test phase the participants heard names of objects they hadseen accompanied by a picture objects that were presented withouta picture and new objects The participants were asked to deter-mine if they had actually seen a picture of the object during thestudy phase We were specifically interested in trials in which theparticipants accurately endorsed perceived pictures as perceived(true memory) and trials in which they inaccurately endorsed imag-ined pictures as perceived (false memory)

The sole purpose of including the lie test was to increase thenumber of false memories available for fMRI analysis which it ap-pears to reliably do In a control experiment conducted outside thescanner half of the studied stimuli were presented in the lie test andhalf were not presented to each participant Overall the false mem-ory rate (ie the rate of inaccurate endorsement of words presentedalone as having been presented along with a picture) during the testphase was significantly higher [t(13) 5 232 p 05] for thoseitems presented in the lie test (17) than for those items that werenot presented in the lie test (13) Although this increases the num-ber of trials available for fMRI data analysis it is nearly impossi-ble to predict what cognitive or neural processes are engaged dur-ing the lie test since true errors made by the participants cannot bedistinguished from deliberate errors Likewise it would be excep-tionally difficult if not impossible to determine in the present par-adigm whether any particular false memory had its source in theoriginal study phase in the lie test or in some combination of bothSince we were interested specifically in examining retrieval pro-cesses yielding false memories and the reconstructive nature of theretrieval process we collected fMRI data only from the test phase

ParticipantsFourteen native English speakers (8 male 6 female) were re-

cruited from the Johns Hopkins University community The partic-ipants were between the ages of 19 and 31 years were right-handedand had no history of neurological or psychiatric illness All theparticipants were naive to the experimental materials and hypothe-ses and gave informed written consent to participate

MaterialsThe stimuli consisted of 450 color photographs of objects on a

white background and auditorily presented names of each object(recorded as sound files by a single female native English speaker)Outside the scanner pictures were presented on a computer screenand the object names were played over headphones Inside thescanner object names were played over pneumatic headphones andresponses were collected using a fiber optic button box

ProcedureThe study phase consisted of a total of 300 trials 150 spoken words

followed by pictures (W1P trials) and 150 spoken words followedby a blank rectangle (W1B trials) The duration of the spoken wordswas approximately 300 msec followed by a delay (~1500 msec)for a total of exactly 1800 msec (see Figure 1) Corresponding pic-tures of the objects or blank rectangles were presented for300 msec followed by a 1500-msec delay The participants wereinstructed to visualize an image of the object and decide if the ob-ject was larger or smaller than a shoebox indicating their responseon the computer keyboard after the word was presented The par-ticipants were informed that a corresponding picture or filler rec-tangle would appear on the screen soon after the word was pre-sented The words presented with pictures and those presented withblank rectangles were counterbalanced In addition to trials withstimuli there were 50 null trials in the form of auditory noise(3600 msec) during which the participant made no response All ofthe stimuli were randomly intermixed The study phase was dividedinto two runs

The lie test consisted of 225 trials (75 of the W1P trials and all150 of the W1B trials from the study phase) of auditorily presentedwords For each word the participants were asked to decide whetheror not they had seen an actual picture of the object in the studyphase On each trial the participants could earn or lose points witheach response with the goal of accumulating as many points as pos-sible (indicated on the screen) Truthful responses were awarded2 6 1 points When the participants lied and endorsed a word stud-ied without a picture they would gain 8 6 1 points on 70 of thetrials and lose 4 6 1 points on 30 of the trials On the latter set oftrials (those checked for accuracy) an animated figure would ap-pear along with a sound effect as the participantrsquos score was re-duced Finally on trials in which the participants lied (or were in-accurate) and did not endorse a trial that had been previouslypresented with a picture they lost 2 6 1 points if the trial waschecked for accuracy but gained 4 6 1 points if it was not checkedTherefore the participants were strongly encouraged to lie and en-dorse words that had not been studied with a picture and they wereonly moderately encouraged to lie and not endorse words that hadbeen studied with a picture The participants were informed thatthey would earn or lose a certain number of points with every re-sponse they made They were given the goal of collecting as manypoints as possible and were told that they would receive feedbackon every trial in the form of a bar graph and the actual number ofpoints collected The participants were informed that although itbenefited them to respond truthfully as to whether or not they hadseen a picture it might benef it them even more to lie if the partic-ular trial was not checked They were told that only a small numberof the trials would be checked The participants indicated their re-sponse on the computer keyboard The trials were self-paced

DOG

300 msec 1500 msec 300 msec 1500 msec

orW+B

W+P

Figure 1 Schematic diagram of the study phase Words (in this example DOG) were pre-sented auditorily followed by a 1500-msec delay Either a picture of the object or a blank rec-tangle was presented for 300 msec followed by a 1500-msec delay The participants imag-ined a picture of the object when the word was presented and a response was required duringthe first 1500-msec delay

fMRI OF FALSE MEMORY 327

The participants were then placed in the fMRI scanner for thetest phase The test phase consisted of 450 trials 150 W1P trials150 W1B trials and 150 trials of spoken words that had not beenpresented at study (foils) Words were presented auditorily every2500 msec The participants were instructed to decide whether ornot they had seen an actual picture of the object in the study phaseThey were instructed that this portion of the experiment pertainedonly to the study phase and to tell the truth to the best of their abil-ity They indicated their ldquoyesrdquo and ldquonordquo responses using a buttonbox In addition there were 75 null trials in the form of auditorynoise (2500 msec) during which the participants made no re-sponse These trials served as a baseline condition for the fMRIdata analysis All of the stimuli were randomly intermixed The testphase was divided into five runs

fMRI MethodsImaging was performed on a Philips Gyroscan 15T MRI scan-

ner equipped with a whole-brain SENSE coil Thirty-f ive T2-weighted triple-oblique functional images were collected per 3-Dvolume using a single-shot echoplanar pulse sequence (64 3 64matrix TE 5 40 msec flip angle 5 90ordm in-plane resolution 5 4 34 mm thickness 5 4 mm TR 5 25 sec) Slices were aligned withthe principal axis of the left and right hippocampus (as determinedby a series of sagittal localizer MRI scans for each participant) Thiswas done to optimize the signal from the medial temporal lobes andto minimize partial-voluming effects so that voxels could be clearlyconstrained to lie within subregions of the MTL A total of 525 vol-umes were collected The task began in synchrony with the acqui-sition of the fifth volume to allow for T1 stabilization After thefunctional scans a high-resolution structural MRI was acquired(MP-RAGE pulse sequence 1 mm3 resolution 150 triple-obliqueaxial slices in the same orientation as the functional images) foranatomical localization

fMRI Data AnalysisImage analysis was performed using Analysis of Functional Neu-

roimages (Cox 1996) Functional MRI data were first resampled intime using a Fourier algorithm to align all slices to a common timebase Functional images were then resampled in space to coregisterthe images and reduce the effects of head motion in three dimen-sions During this process six vectors were created that code for allpossible translations and rotations of the brain fMRI data from alltest runs were concatenated

Following this processing the behavioral data were coded intoeight trial types of interest and a general linear model (GLM) of thefMRI time series data was constructed using these vectors Thestimulus vectors coding for the eight trial types were named ac-cording to study phase condition and participant response as fol-lows (1) SndashS seen pictures endorsed as seen (ldquoyesrdquo response toW1P items) (2) SndashS (no MI) ldquoyesrdquo response to W1P items thatwere not presented during the lie test and therefore received nomisinformation (3) SndashN seen pictures rejected as not seen (ldquonordquoresponse to W1P items) (4) SndashN (no MI) ldquonordquo response to W1Pitems that were not presented during the lie test and therefore re-ceived no misinformation) (5) IndashS imagined pictures endorsed asseen (ldquoyesrdquo response to W1B items) (6) IndashN imagined pictures re-jected as not seen (ldquonordquo response to W1B items) (7) UndashS un-studied words endorsed as seen (ldquoyesrdquo response to foils) and(8) UndashN unstudied words rejected as not seen (ldquonordquo to response tofoils) Unfortunately with only 10 trials per participant on average(range 5 2ndash22) reliable estimates of the hemodynamic response inthe UndashS condition could not be obtained and therefore this trialtype was not included in the group analysis In addition the GLMincluded nuisance vectors coding for first- and second-order driftin the MR signal and for 3-D head motion

The GLM was constructed using a deconvolution technique(Ward 2000) that estimates the impulse response function within

each voxel and performs a multiple linear regression The sum ofthe beta coefficients for the time points corresponding to the ex-pected peak in the hemodynamic response (~25ndash10 sec after stim-ulus onset) was taken as the modelrsquos estimate of the response toeach trial type We should note that this response magnitude is rel-ative to the activity present during the null task (auditory noise) Anactivity level of zero for a particular condition (and a flat hemody-namic response) does not imply a lack of activity in the region dur-ing this condition Rather it implies merely a similar level of ac-tivity in the condition of interest and in the null task which mayvery well be active (Stark amp Squire 2001b)

Initial spatial normalization was done using each participantrsquosstructural MRI to transform data according to the common atlas ofTalairach and Tournoux (1988) This transformation was applied tothe statistical maps of the beta coefficients and in the process thedata were resampled to 25 mm3 The spatially normalized statisti-cal maps of the beta coefficients were blurred using a Gaussian fil-ter with a full-width half maximum of 4 mm to help account forvariations in the functional anatomy across participants

The main analyses involved a voxel-wise two-factor analysis ofvariance (ANOVA) on the beta coefficients with a fixed factor (con-dition all seven trial types) and a random factor (subject) to iden-tify any activation differences between the trial types The ANOVAwas used to define functional regions of interest (ROIs) that demon-strated significant activation differences across any of the seventrial types that passed a voxel-wise threshold of p 01 and a spa-tial extent threshold of 328 mm3 ( p 05 correcting for multiplecomparisons) Post hoc t tests were performed on the average betacoefficients within the ROIs functionally defined by the ANOVAThese t tests were pairwise comparisons of the seven conditionsHemodynamic responses were extracted for three trial types of in-terest (SndashS IndashS and IndashN) for the identif ied regions that showedsignificant differences among these three conditions

The MTL analysis was performed using the ROIndashAL (regions ofinterestndashalignment) technique described by Stark and Okado (2003)The technique begins with a definition of the structures in the MTL(hippocampal region temporopolar perirhinal entorhinal andparahippocampal cortices) bilaterally according to the techniquesdescribed by Insausti et al (1998) The parahippocampal cortexwas further def ined bilaterally as the portion of the parahippo-campal gyrus caudal to the perirhinal cortex and rostral to the sple-nium of the corpus callosum (by this definition the rostral extentincludes only slightly more tissue than as def ined by Pruessneret al 2002) The hippocampal region (the CA fields of the hip-pocampus dentate gyrus and subiculum) was also defined bilater-ally ROIndashAL uses these anatomically defined ROIs to calculate anadditional transformation matrix to fine-tune the cross-participantalignment in the MTL ROIndashAL significantly improves the overlapacross participants increasing statistical power and precision in lo-calization of cross-participant tests In the ROIndashAL analysis avoxel-wise threshold of p 01 and a spatial extent threshold of125 mm3 was used ( p 05 correcting for multiple comparisonswith the MTL alone)

RESULTS

Behavioral ResultsThe overall response rates for each category are shown

in Figure 2a To assess accuracy and false memory ratesoverall discriminability (d cent) scores were calculated fromthe SndashS rate for each of the two types of false alarmrates The dcent for SndashS and IndashS was 138 and the dcent forSndashS and UndashS items was 198 These d cent scores signifi-cantly differed [t(13) 5 534 p 001] as did the over-all rates for the two false alarm conditions [t(13) 5 343

328 OKADO AND STARK

p 01] suggesting that the combination of the realitymonitoring paradigm and the lie test significantly in-creased the false alarm rate to above chance levels andthat the participants were actually experiencing falsememories During the lie test the participants claimed476 of W1B items were studied as pictures Fifty-three percent of the IndashS items from the test phase wereitems that were falsely endorsed during the lie test

Of the seven possible trial types three critical trialtypes were selected a priori for further analyses in an ef-fort to analyze differences between true and false mem-ories SndashS IndashS and IndashN (the left three bars in Figures 2and 3cndashe) Differences between SndashS and IndashS activationsare suggestive of differences between true and falsememory retrieval Both trial types were presented in thestudy phase (SndashS as W1P and IndashS as W1B) and in thelie test and both were endorsed in the test phase (SndashScorrectly endorsed and IndashS incorrectly endorsed) Simi-larly a contrast between IndashS and IndashN can highlight dif-ferences between true and false memory retrieval by in-dicating differences between the false endorsement of animagined picture as being real and the correct rejectionof an imagined picture as having been only imagined (inboth of these conditions the stimuli have been treatedidentically prior to the test phase) Finally by contrast-ing true retrieval of pictures and correct rejection of onlyimagined pictures differences between SndashS and IndashN ac-tivations are suggestive of accurate retrieval However itshould be noted that although this contrast is often usedto assess memory retrieval success it can be contami-nated and weakened by activity associated with inciden-tal encoding during the retrieval task (Stark amp Okado2003)

Figure 2b shows the mean reaction times (RT) for alltrial types In the three conditions of interest the mean

RTs were fastest for SndashS items (1313 msec) and slow-est for IndashS items (1469 msec) A repeated measuresANOVA showed that there was a significant effect ofthese three trial types on RT [F(113) 5 225 p 001]Pairwise t tests showed significant RT differences betweenSndashS and IndashS [156 msec t(13) 5 2474 p 01] be-tween IndashS and IndashN [95 msec t(13) 5 271 p 05] andbetween SndashS and IndashN [61 msec t(13) 5 2306 p 01]

f MRI ResultsA voxel-wise two-factor ANOVA was first conducted

on the whole brain data to identify regions that showedany activation differences among the seven trial typesResults of this analysis identified 12 areas in which ac-tivity significantly differed across trial types (Figure 3a)These areas were treated as functionally defined ROIsand the average hemodynamic response functions foreach trial type were calculated for each region

The 12 regions identified demonstrated three distinctpatterns of activity in the functional ROI analysis withrespect to our three conditions of interest The first no-table pattern observed in the identified regions was sim-ilar increased activity for SndashS and IndashS in comparisonwith IndashN (Figure 3c) This pattern of activity was ob-served in the left parietal cortex and left frontal regionsRegions in which SndashS and IndashN differed significantly andIndashS and IndashN differed significantly included the left pari-etal cortex [BA 73940 t(13) 5 329 p 01 andt(13) 5 257 p 05 respectively] left precentral gyrus[BA 9 t(13) 5 245 p 05 and t(13) 5 319 p 01respectively] and left inferior frontal gyrus [BA 1046t(13) 5 261 p 05 and t(13) 5 298 p 05 respec-tively] The left caudate showed this similar pattern ofactivation with SndashS and IndashS demonstrating increasedactivity in comparison with IndashN but showed no signifi-

Figure 2 Behavioral data from the test phase Percentage of trials endorsed as having beenpresented with a picture at time of study (a) and reaction time for the decision (b) are plot-ted for each trial type as means across all 14 participants Error bars indicate the standarderrors of the means

fMRI OF FALSE MEMORY 329

Figure 3 (a and b) Locations and sample hemodynamic responses for regions where activity at testvaried among the seven trial types analyzed (there were too few trials in the UndashS condition to ana-lyze) (a) The 12 regions where activity varied as a function of trial type are shown as colored overlayson 3-D renderings of a brain A left parietal cortex (BA 73940) B left precentral gyrus (BA 9) Cleft inferior frontal gyrus (BA 1046) D left caudate E right middle occipital gyrus (BA 1819) Fleft cuneus (BA 171819) G left lingual gyrus (BA 19) H left fusiform gyrus (1819) I bilaterallingual gyrus (BA 18) J bilateral cuneus (BA 17182730) K right posterior parahippocampalgyrus (BA 30353637) L right anterior cingulate gyrus (BA 6932) (b) Average hemodynamic re-sponse functions (sum of beta coefficients vs image acquisition [TR]) for the three trial types of in-terest (SndashS IndashS and IndashN) are shown for three sample regions (left precentral gyrus left cuneus andright anterior cingulate) that demonstrate the patterns of activity observed (cndashe) Activity for all seventrial types analyzed in all 12 functionally defined ROIs Bars show the mean fMRI response (sum ofbeta coefficients) across participants and error bars show the standard errors of the means Aster-isks indicate significant differences in activity between conditions of interest (SndashS IndashS and IndashN) (c)Regions demonstrating the first pattern of activity included A left parietal cortex B left precentralgyrus C left inferior frontal gyrus and D left caudate (d) Regions demonstrating the second pat-tern of activity included E right middle occipital gyrus F left cuneus G left lingual gyrus H leftfusiform gyrus I bilateral lingual gyrus J bilateral cuneus and K right posterior parahippocampalgyrus (e) The third pattern of activity was observed in L right anterior cingulate gyrus

B L precentralgyrus (BA 9)

F L cuneus(BA 171819)

L R anteriorcingulate (BA 6932)

TR (25 sec)

beta

wei

ghts

b

c

d

050

ndash05ndash1

ndash15

08

04

0

05

0

ndash05

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

4

2

0

2

2

0

ndash2

ndash4

2

0

ndash2

ndash4

4

2

0

ndash2

A B C D

E F G H

I J K e L

S-S

I-S

I-N

S-S(No MI)

S-N

S-N(No MI)

U-N

1 2 3 4 5 6 7 8

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

326 OKADO AND STARK

1974) the participantsrsquo memories appear to have been altered to thepoint that they falsely believed they had actually seen the picturesince the false alarm rate was raised significantly

The test phase was the only in-scanner portion of the experimentIn the test phase the participants heard names of objects they hadseen accompanied by a picture objects that were presented withouta picture and new objects The participants were asked to deter-mine if they had actually seen a picture of the object during thestudy phase We were specifically interested in trials in which theparticipants accurately endorsed perceived pictures as perceived(true memory) and trials in which they inaccurately endorsed imag-ined pictures as perceived (false memory)

The sole purpose of including the lie test was to increase thenumber of false memories available for fMRI analysis which it ap-pears to reliably do In a control experiment conducted outside thescanner half of the studied stimuli were presented in the lie test andhalf were not presented to each participant Overall the false mem-ory rate (ie the rate of inaccurate endorsement of words presentedalone as having been presented along with a picture) during the testphase was significantly higher [t(13) 5 232 p 05] for thoseitems presented in the lie test (17) than for those items that werenot presented in the lie test (13) Although this increases the num-ber of trials available for fMRI data analysis it is nearly impossi-ble to predict what cognitive or neural processes are engaged dur-ing the lie test since true errors made by the participants cannot bedistinguished from deliberate errors Likewise it would be excep-tionally difficult if not impossible to determine in the present par-adigm whether any particular false memory had its source in theoriginal study phase in the lie test or in some combination of bothSince we were interested specifically in examining retrieval pro-cesses yielding false memories and the reconstructive nature of theretrieval process we collected fMRI data only from the test phase

ParticipantsFourteen native English speakers (8 male 6 female) were re-

cruited from the Johns Hopkins University community The partic-ipants were between the ages of 19 and 31 years were right-handedand had no history of neurological or psychiatric illness All theparticipants were naive to the experimental materials and hypothe-ses and gave informed written consent to participate

MaterialsThe stimuli consisted of 450 color photographs of objects on a

white background and auditorily presented names of each object(recorded as sound files by a single female native English speaker)Outside the scanner pictures were presented on a computer screenand the object names were played over headphones Inside thescanner object names were played over pneumatic headphones andresponses were collected using a fiber optic button box

ProcedureThe study phase consisted of a total of 300 trials 150 spoken words

followed by pictures (W1P trials) and 150 spoken words followedby a blank rectangle (W1B trials) The duration of the spoken wordswas approximately 300 msec followed by a delay (~1500 msec)for a total of exactly 1800 msec (see Figure 1) Corresponding pic-tures of the objects or blank rectangles were presented for300 msec followed by a 1500-msec delay The participants wereinstructed to visualize an image of the object and decide if the ob-ject was larger or smaller than a shoebox indicating their responseon the computer keyboard after the word was presented The par-ticipants were informed that a corresponding picture or filler rec-tangle would appear on the screen soon after the word was pre-sented The words presented with pictures and those presented withblank rectangles were counterbalanced In addition to trials withstimuli there were 50 null trials in the form of auditory noise(3600 msec) during which the participant made no response All ofthe stimuli were randomly intermixed The study phase was dividedinto two runs

The lie test consisted of 225 trials (75 of the W1P trials and all150 of the W1B trials from the study phase) of auditorily presentedwords For each word the participants were asked to decide whetheror not they had seen an actual picture of the object in the studyphase On each trial the participants could earn or lose points witheach response with the goal of accumulating as many points as pos-sible (indicated on the screen) Truthful responses were awarded2 6 1 points When the participants lied and endorsed a word stud-ied without a picture they would gain 8 6 1 points on 70 of thetrials and lose 4 6 1 points on 30 of the trials On the latter set oftrials (those checked for accuracy) an animated figure would ap-pear along with a sound effect as the participantrsquos score was re-duced Finally on trials in which the participants lied (or were in-accurate) and did not endorse a trial that had been previouslypresented with a picture they lost 2 6 1 points if the trial waschecked for accuracy but gained 4 6 1 points if it was not checkedTherefore the participants were strongly encouraged to lie and en-dorse words that had not been studied with a picture and they wereonly moderately encouraged to lie and not endorse words that hadbeen studied with a picture The participants were informed thatthey would earn or lose a certain number of points with every re-sponse they made They were given the goal of collecting as manypoints as possible and were told that they would receive feedbackon every trial in the form of a bar graph and the actual number ofpoints collected The participants were informed that although itbenefited them to respond truthfully as to whether or not they hadseen a picture it might benef it them even more to lie if the partic-ular trial was not checked They were told that only a small numberof the trials would be checked The participants indicated their re-sponse on the computer keyboard The trials were self-paced

DOG

300 msec 1500 msec 300 msec 1500 msec

orW+B

W+P

Figure 1 Schematic diagram of the study phase Words (in this example DOG) were pre-sented auditorily followed by a 1500-msec delay Either a picture of the object or a blank rec-tangle was presented for 300 msec followed by a 1500-msec delay The participants imag-ined a picture of the object when the word was presented and a response was required duringthe first 1500-msec delay

fMRI OF FALSE MEMORY 327

The participants were then placed in the fMRI scanner for thetest phase The test phase consisted of 450 trials 150 W1P trials150 W1B trials and 150 trials of spoken words that had not beenpresented at study (foils) Words were presented auditorily every2500 msec The participants were instructed to decide whether ornot they had seen an actual picture of the object in the study phaseThey were instructed that this portion of the experiment pertainedonly to the study phase and to tell the truth to the best of their abil-ity They indicated their ldquoyesrdquo and ldquonordquo responses using a buttonbox In addition there were 75 null trials in the form of auditorynoise (2500 msec) during which the participants made no re-sponse These trials served as a baseline condition for the fMRIdata analysis All of the stimuli were randomly intermixed The testphase was divided into five runs

fMRI MethodsImaging was performed on a Philips Gyroscan 15T MRI scan-

ner equipped with a whole-brain SENSE coil Thirty-f ive T2-weighted triple-oblique functional images were collected per 3-Dvolume using a single-shot echoplanar pulse sequence (64 3 64matrix TE 5 40 msec flip angle 5 90ordm in-plane resolution 5 4 34 mm thickness 5 4 mm TR 5 25 sec) Slices were aligned withthe principal axis of the left and right hippocampus (as determinedby a series of sagittal localizer MRI scans for each participant) Thiswas done to optimize the signal from the medial temporal lobes andto minimize partial-voluming effects so that voxels could be clearlyconstrained to lie within subregions of the MTL A total of 525 vol-umes were collected The task began in synchrony with the acqui-sition of the fifth volume to allow for T1 stabilization After thefunctional scans a high-resolution structural MRI was acquired(MP-RAGE pulse sequence 1 mm3 resolution 150 triple-obliqueaxial slices in the same orientation as the functional images) foranatomical localization

fMRI Data AnalysisImage analysis was performed using Analysis of Functional Neu-

roimages (Cox 1996) Functional MRI data were first resampled intime using a Fourier algorithm to align all slices to a common timebase Functional images were then resampled in space to coregisterthe images and reduce the effects of head motion in three dimen-sions During this process six vectors were created that code for allpossible translations and rotations of the brain fMRI data from alltest runs were concatenated

Following this processing the behavioral data were coded intoeight trial types of interest and a general linear model (GLM) of thefMRI time series data was constructed using these vectors Thestimulus vectors coding for the eight trial types were named ac-cording to study phase condition and participant response as fol-lows (1) SndashS seen pictures endorsed as seen (ldquoyesrdquo response toW1P items) (2) SndashS (no MI) ldquoyesrdquo response to W1P items thatwere not presented during the lie test and therefore received nomisinformation (3) SndashN seen pictures rejected as not seen (ldquonordquoresponse to W1P items) (4) SndashN (no MI) ldquonordquo response to W1Pitems that were not presented during the lie test and therefore re-ceived no misinformation) (5) IndashS imagined pictures endorsed asseen (ldquoyesrdquo response to W1B items) (6) IndashN imagined pictures re-jected as not seen (ldquonordquo response to W1B items) (7) UndashS un-studied words endorsed as seen (ldquoyesrdquo response to foils) and(8) UndashN unstudied words rejected as not seen (ldquonordquo to response tofoils) Unfortunately with only 10 trials per participant on average(range 5 2ndash22) reliable estimates of the hemodynamic response inthe UndashS condition could not be obtained and therefore this trialtype was not included in the group analysis In addition the GLMincluded nuisance vectors coding for first- and second-order driftin the MR signal and for 3-D head motion

The GLM was constructed using a deconvolution technique(Ward 2000) that estimates the impulse response function within

each voxel and performs a multiple linear regression The sum ofthe beta coefficients for the time points corresponding to the ex-pected peak in the hemodynamic response (~25ndash10 sec after stim-ulus onset) was taken as the modelrsquos estimate of the response toeach trial type We should note that this response magnitude is rel-ative to the activity present during the null task (auditory noise) Anactivity level of zero for a particular condition (and a flat hemody-namic response) does not imply a lack of activity in the region dur-ing this condition Rather it implies merely a similar level of ac-tivity in the condition of interest and in the null task which mayvery well be active (Stark amp Squire 2001b)

Initial spatial normalization was done using each participantrsquosstructural MRI to transform data according to the common atlas ofTalairach and Tournoux (1988) This transformation was applied tothe statistical maps of the beta coefficients and in the process thedata were resampled to 25 mm3 The spatially normalized statisti-cal maps of the beta coefficients were blurred using a Gaussian fil-ter with a full-width half maximum of 4 mm to help account forvariations in the functional anatomy across participants

The main analyses involved a voxel-wise two-factor analysis ofvariance (ANOVA) on the beta coefficients with a fixed factor (con-dition all seven trial types) and a random factor (subject) to iden-tify any activation differences between the trial types The ANOVAwas used to define functional regions of interest (ROIs) that demon-strated significant activation differences across any of the seventrial types that passed a voxel-wise threshold of p 01 and a spa-tial extent threshold of 328 mm3 ( p 05 correcting for multiplecomparisons) Post hoc t tests were performed on the average betacoefficients within the ROIs functionally defined by the ANOVAThese t tests were pairwise comparisons of the seven conditionsHemodynamic responses were extracted for three trial types of in-terest (SndashS IndashS and IndashN) for the identif ied regions that showedsignificant differences among these three conditions

The MTL analysis was performed using the ROIndashAL (regions ofinterestndashalignment) technique described by Stark and Okado (2003)The technique begins with a definition of the structures in the MTL(hippocampal region temporopolar perirhinal entorhinal andparahippocampal cortices) bilaterally according to the techniquesdescribed by Insausti et al (1998) The parahippocampal cortexwas further def ined bilaterally as the portion of the parahippo-campal gyrus caudal to the perirhinal cortex and rostral to the sple-nium of the corpus callosum (by this definition the rostral extentincludes only slightly more tissue than as def ined by Pruessneret al 2002) The hippocampal region (the CA fields of the hip-pocampus dentate gyrus and subiculum) was also defined bilater-ally ROIndashAL uses these anatomically defined ROIs to calculate anadditional transformation matrix to fine-tune the cross-participantalignment in the MTL ROIndashAL significantly improves the overlapacross participants increasing statistical power and precision in lo-calization of cross-participant tests In the ROIndashAL analysis avoxel-wise threshold of p 01 and a spatial extent threshold of125 mm3 was used ( p 05 correcting for multiple comparisonswith the MTL alone)

RESULTS

Behavioral ResultsThe overall response rates for each category are shown

in Figure 2a To assess accuracy and false memory ratesoverall discriminability (d cent) scores were calculated fromthe SndashS rate for each of the two types of false alarmrates The dcent for SndashS and IndashS was 138 and the dcent forSndashS and UndashS items was 198 These d cent scores signifi-cantly differed [t(13) 5 534 p 001] as did the over-all rates for the two false alarm conditions [t(13) 5 343

328 OKADO AND STARK

p 01] suggesting that the combination of the realitymonitoring paradigm and the lie test significantly in-creased the false alarm rate to above chance levels andthat the participants were actually experiencing falsememories During the lie test the participants claimed476 of W1B items were studied as pictures Fifty-three percent of the IndashS items from the test phase wereitems that were falsely endorsed during the lie test

Of the seven possible trial types three critical trialtypes were selected a priori for further analyses in an ef-fort to analyze differences between true and false mem-ories SndashS IndashS and IndashN (the left three bars in Figures 2and 3cndashe) Differences between SndashS and IndashS activationsare suggestive of differences between true and falsememory retrieval Both trial types were presented in thestudy phase (SndashS as W1P and IndashS as W1B) and in thelie test and both were endorsed in the test phase (SndashScorrectly endorsed and IndashS incorrectly endorsed) Simi-larly a contrast between IndashS and IndashN can highlight dif-ferences between true and false memory retrieval by in-dicating differences between the false endorsement of animagined picture as being real and the correct rejectionof an imagined picture as having been only imagined (inboth of these conditions the stimuli have been treatedidentically prior to the test phase) Finally by contrast-ing true retrieval of pictures and correct rejection of onlyimagined pictures differences between SndashS and IndashN ac-tivations are suggestive of accurate retrieval However itshould be noted that although this contrast is often usedto assess memory retrieval success it can be contami-nated and weakened by activity associated with inciden-tal encoding during the retrieval task (Stark amp Okado2003)

Figure 2b shows the mean reaction times (RT) for alltrial types In the three conditions of interest the mean

RTs were fastest for SndashS items (1313 msec) and slow-est for IndashS items (1469 msec) A repeated measuresANOVA showed that there was a significant effect ofthese three trial types on RT [F(113) 5 225 p 001]Pairwise t tests showed significant RT differences betweenSndashS and IndashS [156 msec t(13) 5 2474 p 01] be-tween IndashS and IndashN [95 msec t(13) 5 271 p 05] andbetween SndashS and IndashN [61 msec t(13) 5 2306 p 01]

f MRI ResultsA voxel-wise two-factor ANOVA was first conducted

on the whole brain data to identify regions that showedany activation differences among the seven trial typesResults of this analysis identified 12 areas in which ac-tivity significantly differed across trial types (Figure 3a)These areas were treated as functionally defined ROIsand the average hemodynamic response functions foreach trial type were calculated for each region

The 12 regions identified demonstrated three distinctpatterns of activity in the functional ROI analysis withrespect to our three conditions of interest The first no-table pattern observed in the identified regions was sim-ilar increased activity for SndashS and IndashS in comparisonwith IndashN (Figure 3c) This pattern of activity was ob-served in the left parietal cortex and left frontal regionsRegions in which SndashS and IndashN differed significantly andIndashS and IndashN differed significantly included the left pari-etal cortex [BA 73940 t(13) 5 329 p 01 andt(13) 5 257 p 05 respectively] left precentral gyrus[BA 9 t(13) 5 245 p 05 and t(13) 5 319 p 01respectively] and left inferior frontal gyrus [BA 1046t(13) 5 261 p 05 and t(13) 5 298 p 05 respec-tively] The left caudate showed this similar pattern ofactivation with SndashS and IndashS demonstrating increasedactivity in comparison with IndashN but showed no signifi-

Figure 2 Behavioral data from the test phase Percentage of trials endorsed as having beenpresented with a picture at time of study (a) and reaction time for the decision (b) are plot-ted for each trial type as means across all 14 participants Error bars indicate the standarderrors of the means

fMRI OF FALSE MEMORY 329

Figure 3 (a and b) Locations and sample hemodynamic responses for regions where activity at testvaried among the seven trial types analyzed (there were too few trials in the UndashS condition to ana-lyze) (a) The 12 regions where activity varied as a function of trial type are shown as colored overlayson 3-D renderings of a brain A left parietal cortex (BA 73940) B left precentral gyrus (BA 9) Cleft inferior frontal gyrus (BA 1046) D left caudate E right middle occipital gyrus (BA 1819) Fleft cuneus (BA 171819) G left lingual gyrus (BA 19) H left fusiform gyrus (1819) I bilaterallingual gyrus (BA 18) J bilateral cuneus (BA 17182730) K right posterior parahippocampalgyrus (BA 30353637) L right anterior cingulate gyrus (BA 6932) (b) Average hemodynamic re-sponse functions (sum of beta coefficients vs image acquisition [TR]) for the three trial types of in-terest (SndashS IndashS and IndashN) are shown for three sample regions (left precentral gyrus left cuneus andright anterior cingulate) that demonstrate the patterns of activity observed (cndashe) Activity for all seventrial types analyzed in all 12 functionally defined ROIs Bars show the mean fMRI response (sum ofbeta coefficients) across participants and error bars show the standard errors of the means Aster-isks indicate significant differences in activity between conditions of interest (SndashS IndashS and IndashN) (c)Regions demonstrating the first pattern of activity included A left parietal cortex B left precentralgyrus C left inferior frontal gyrus and D left caudate (d) Regions demonstrating the second pat-tern of activity included E right middle occipital gyrus F left cuneus G left lingual gyrus H leftfusiform gyrus I bilateral lingual gyrus J bilateral cuneus and K right posterior parahippocampalgyrus (e) The third pattern of activity was observed in L right anterior cingulate gyrus

B L precentralgyrus (BA 9)

F L cuneus(BA 171819)

L R anteriorcingulate (BA 6932)

TR (25 sec)

beta

wei

ghts

b

c

d

050

ndash05ndash1

ndash15

08

04

0

05

0

ndash05

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

4

2

0

2

2

0

ndash2

ndash4

2

0

ndash2

ndash4

4

2

0

ndash2

A B C D

E F G H

I J K e L

S-S

I-S

I-N

S-S(No MI)

S-N

S-N(No MI)

U-N

1 2 3 4 5 6 7 8

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

fMRI OF FALSE MEMORY 327

The participants were then placed in the fMRI scanner for thetest phase The test phase consisted of 450 trials 150 W1P trials150 W1B trials and 150 trials of spoken words that had not beenpresented at study (foils) Words were presented auditorily every2500 msec The participants were instructed to decide whether ornot they had seen an actual picture of the object in the study phaseThey were instructed that this portion of the experiment pertainedonly to the study phase and to tell the truth to the best of their abil-ity They indicated their ldquoyesrdquo and ldquonordquo responses using a buttonbox In addition there were 75 null trials in the form of auditorynoise (2500 msec) during which the participants made no re-sponse These trials served as a baseline condition for the fMRIdata analysis All of the stimuli were randomly intermixed The testphase was divided into five runs

fMRI MethodsImaging was performed on a Philips Gyroscan 15T MRI scan-

ner equipped with a whole-brain SENSE coil Thirty-f ive T2-weighted triple-oblique functional images were collected per 3-Dvolume using a single-shot echoplanar pulse sequence (64 3 64matrix TE 5 40 msec flip angle 5 90ordm in-plane resolution 5 4 34 mm thickness 5 4 mm TR 5 25 sec) Slices were aligned withthe principal axis of the left and right hippocampus (as determinedby a series of sagittal localizer MRI scans for each participant) Thiswas done to optimize the signal from the medial temporal lobes andto minimize partial-voluming effects so that voxels could be clearlyconstrained to lie within subregions of the MTL A total of 525 vol-umes were collected The task began in synchrony with the acqui-sition of the fifth volume to allow for T1 stabilization After thefunctional scans a high-resolution structural MRI was acquired(MP-RAGE pulse sequence 1 mm3 resolution 150 triple-obliqueaxial slices in the same orientation as the functional images) foranatomical localization

fMRI Data AnalysisImage analysis was performed using Analysis of Functional Neu-

roimages (Cox 1996) Functional MRI data were first resampled intime using a Fourier algorithm to align all slices to a common timebase Functional images were then resampled in space to coregisterthe images and reduce the effects of head motion in three dimen-sions During this process six vectors were created that code for allpossible translations and rotations of the brain fMRI data from alltest runs were concatenated

Following this processing the behavioral data were coded intoeight trial types of interest and a general linear model (GLM) of thefMRI time series data was constructed using these vectors Thestimulus vectors coding for the eight trial types were named ac-cording to study phase condition and participant response as fol-lows (1) SndashS seen pictures endorsed as seen (ldquoyesrdquo response toW1P items) (2) SndashS (no MI) ldquoyesrdquo response to W1P items thatwere not presented during the lie test and therefore received nomisinformation (3) SndashN seen pictures rejected as not seen (ldquonordquoresponse to W1P items) (4) SndashN (no MI) ldquonordquo response to W1Pitems that were not presented during the lie test and therefore re-ceived no misinformation) (5) IndashS imagined pictures endorsed asseen (ldquoyesrdquo response to W1B items) (6) IndashN imagined pictures re-jected as not seen (ldquonordquo response to W1B items) (7) UndashS un-studied words endorsed as seen (ldquoyesrdquo response to foils) and(8) UndashN unstudied words rejected as not seen (ldquonordquo to response tofoils) Unfortunately with only 10 trials per participant on average(range 5 2ndash22) reliable estimates of the hemodynamic response inthe UndashS condition could not be obtained and therefore this trialtype was not included in the group analysis In addition the GLMincluded nuisance vectors coding for first- and second-order driftin the MR signal and for 3-D head motion

The GLM was constructed using a deconvolution technique(Ward 2000) that estimates the impulse response function within

each voxel and performs a multiple linear regression The sum ofthe beta coefficients for the time points corresponding to the ex-pected peak in the hemodynamic response (~25ndash10 sec after stim-ulus onset) was taken as the modelrsquos estimate of the response toeach trial type We should note that this response magnitude is rel-ative to the activity present during the null task (auditory noise) Anactivity level of zero for a particular condition (and a flat hemody-namic response) does not imply a lack of activity in the region dur-ing this condition Rather it implies merely a similar level of ac-tivity in the condition of interest and in the null task which mayvery well be active (Stark amp Squire 2001b)

Initial spatial normalization was done using each participantrsquosstructural MRI to transform data according to the common atlas ofTalairach and Tournoux (1988) This transformation was applied tothe statistical maps of the beta coefficients and in the process thedata were resampled to 25 mm3 The spatially normalized statisti-cal maps of the beta coefficients were blurred using a Gaussian fil-ter with a full-width half maximum of 4 mm to help account forvariations in the functional anatomy across participants

The main analyses involved a voxel-wise two-factor analysis ofvariance (ANOVA) on the beta coefficients with a fixed factor (con-dition all seven trial types) and a random factor (subject) to iden-tify any activation differences between the trial types The ANOVAwas used to define functional regions of interest (ROIs) that demon-strated significant activation differences across any of the seventrial types that passed a voxel-wise threshold of p 01 and a spa-tial extent threshold of 328 mm3 ( p 05 correcting for multiplecomparisons) Post hoc t tests were performed on the average betacoefficients within the ROIs functionally defined by the ANOVAThese t tests were pairwise comparisons of the seven conditionsHemodynamic responses were extracted for three trial types of in-terest (SndashS IndashS and IndashN) for the identif ied regions that showedsignificant differences among these three conditions

The MTL analysis was performed using the ROIndashAL (regions ofinterestndashalignment) technique described by Stark and Okado (2003)The technique begins with a definition of the structures in the MTL(hippocampal region temporopolar perirhinal entorhinal andparahippocampal cortices) bilaterally according to the techniquesdescribed by Insausti et al (1998) The parahippocampal cortexwas further def ined bilaterally as the portion of the parahippo-campal gyrus caudal to the perirhinal cortex and rostral to the sple-nium of the corpus callosum (by this definition the rostral extentincludes only slightly more tissue than as def ined by Pruessneret al 2002) The hippocampal region (the CA fields of the hip-pocampus dentate gyrus and subiculum) was also defined bilater-ally ROIndashAL uses these anatomically defined ROIs to calculate anadditional transformation matrix to fine-tune the cross-participantalignment in the MTL ROIndashAL significantly improves the overlapacross participants increasing statistical power and precision in lo-calization of cross-participant tests In the ROIndashAL analysis avoxel-wise threshold of p 01 and a spatial extent threshold of125 mm3 was used ( p 05 correcting for multiple comparisonswith the MTL alone)

RESULTS

Behavioral ResultsThe overall response rates for each category are shown

in Figure 2a To assess accuracy and false memory ratesoverall discriminability (d cent) scores were calculated fromthe SndashS rate for each of the two types of false alarmrates The dcent for SndashS and IndashS was 138 and the dcent forSndashS and UndashS items was 198 These d cent scores signifi-cantly differed [t(13) 5 534 p 001] as did the over-all rates for the two false alarm conditions [t(13) 5 343

328 OKADO AND STARK

p 01] suggesting that the combination of the realitymonitoring paradigm and the lie test significantly in-creased the false alarm rate to above chance levels andthat the participants were actually experiencing falsememories During the lie test the participants claimed476 of W1B items were studied as pictures Fifty-three percent of the IndashS items from the test phase wereitems that were falsely endorsed during the lie test

Of the seven possible trial types three critical trialtypes were selected a priori for further analyses in an ef-fort to analyze differences between true and false mem-ories SndashS IndashS and IndashN (the left three bars in Figures 2and 3cndashe) Differences between SndashS and IndashS activationsare suggestive of differences between true and falsememory retrieval Both trial types were presented in thestudy phase (SndashS as W1P and IndashS as W1B) and in thelie test and both were endorsed in the test phase (SndashScorrectly endorsed and IndashS incorrectly endorsed) Simi-larly a contrast between IndashS and IndashN can highlight dif-ferences between true and false memory retrieval by in-dicating differences between the false endorsement of animagined picture as being real and the correct rejectionof an imagined picture as having been only imagined (inboth of these conditions the stimuli have been treatedidentically prior to the test phase) Finally by contrast-ing true retrieval of pictures and correct rejection of onlyimagined pictures differences between SndashS and IndashN ac-tivations are suggestive of accurate retrieval However itshould be noted that although this contrast is often usedto assess memory retrieval success it can be contami-nated and weakened by activity associated with inciden-tal encoding during the retrieval task (Stark amp Okado2003)

Figure 2b shows the mean reaction times (RT) for alltrial types In the three conditions of interest the mean

RTs were fastest for SndashS items (1313 msec) and slow-est for IndashS items (1469 msec) A repeated measuresANOVA showed that there was a significant effect ofthese three trial types on RT [F(113) 5 225 p 001]Pairwise t tests showed significant RT differences betweenSndashS and IndashS [156 msec t(13) 5 2474 p 01] be-tween IndashS and IndashN [95 msec t(13) 5 271 p 05] andbetween SndashS and IndashN [61 msec t(13) 5 2306 p 01]

f MRI ResultsA voxel-wise two-factor ANOVA was first conducted

on the whole brain data to identify regions that showedany activation differences among the seven trial typesResults of this analysis identified 12 areas in which ac-tivity significantly differed across trial types (Figure 3a)These areas were treated as functionally defined ROIsand the average hemodynamic response functions foreach trial type were calculated for each region

The 12 regions identified demonstrated three distinctpatterns of activity in the functional ROI analysis withrespect to our three conditions of interest The first no-table pattern observed in the identified regions was sim-ilar increased activity for SndashS and IndashS in comparisonwith IndashN (Figure 3c) This pattern of activity was ob-served in the left parietal cortex and left frontal regionsRegions in which SndashS and IndashN differed significantly andIndashS and IndashN differed significantly included the left pari-etal cortex [BA 73940 t(13) 5 329 p 01 andt(13) 5 257 p 05 respectively] left precentral gyrus[BA 9 t(13) 5 245 p 05 and t(13) 5 319 p 01respectively] and left inferior frontal gyrus [BA 1046t(13) 5 261 p 05 and t(13) 5 298 p 05 respec-tively] The left caudate showed this similar pattern ofactivation with SndashS and IndashS demonstrating increasedactivity in comparison with IndashN but showed no signifi-

Figure 2 Behavioral data from the test phase Percentage of trials endorsed as having beenpresented with a picture at time of study (a) and reaction time for the decision (b) are plot-ted for each trial type as means across all 14 participants Error bars indicate the standarderrors of the means

fMRI OF FALSE MEMORY 329

Figure 3 (a and b) Locations and sample hemodynamic responses for regions where activity at testvaried among the seven trial types analyzed (there were too few trials in the UndashS condition to ana-lyze) (a) The 12 regions where activity varied as a function of trial type are shown as colored overlayson 3-D renderings of a brain A left parietal cortex (BA 73940) B left precentral gyrus (BA 9) Cleft inferior frontal gyrus (BA 1046) D left caudate E right middle occipital gyrus (BA 1819) Fleft cuneus (BA 171819) G left lingual gyrus (BA 19) H left fusiform gyrus (1819) I bilaterallingual gyrus (BA 18) J bilateral cuneus (BA 17182730) K right posterior parahippocampalgyrus (BA 30353637) L right anterior cingulate gyrus (BA 6932) (b) Average hemodynamic re-sponse functions (sum of beta coefficients vs image acquisition [TR]) for the three trial types of in-terest (SndashS IndashS and IndashN) are shown for three sample regions (left precentral gyrus left cuneus andright anterior cingulate) that demonstrate the patterns of activity observed (cndashe) Activity for all seventrial types analyzed in all 12 functionally defined ROIs Bars show the mean fMRI response (sum ofbeta coefficients) across participants and error bars show the standard errors of the means Aster-isks indicate significant differences in activity between conditions of interest (SndashS IndashS and IndashN) (c)Regions demonstrating the first pattern of activity included A left parietal cortex B left precentralgyrus C left inferior frontal gyrus and D left caudate (d) Regions demonstrating the second pat-tern of activity included E right middle occipital gyrus F left cuneus G left lingual gyrus H leftfusiform gyrus I bilateral lingual gyrus J bilateral cuneus and K right posterior parahippocampalgyrus (e) The third pattern of activity was observed in L right anterior cingulate gyrus

B L precentralgyrus (BA 9)

F L cuneus(BA 171819)

L R anteriorcingulate (BA 6932)

TR (25 sec)

beta

wei

ghts

b

c

d

050

ndash05ndash1

ndash15

08

04

0

05

0

ndash05

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

4

2

0

2

2

0

ndash2

ndash4

2

0

ndash2

ndash4

4

2

0

ndash2

A B C D

E F G H

I J K e L

S-S

I-S

I-N

S-S(No MI)

S-N

S-N(No MI)

U-N

1 2 3 4 5 6 7 8

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

328 OKADO AND STARK

p 01] suggesting that the combination of the realitymonitoring paradigm and the lie test significantly in-creased the false alarm rate to above chance levels andthat the participants were actually experiencing falsememories During the lie test the participants claimed476 of W1B items were studied as pictures Fifty-three percent of the IndashS items from the test phase wereitems that were falsely endorsed during the lie test

Of the seven possible trial types three critical trialtypes were selected a priori for further analyses in an ef-fort to analyze differences between true and false mem-ories SndashS IndashS and IndashN (the left three bars in Figures 2and 3cndashe) Differences between SndashS and IndashS activationsare suggestive of differences between true and falsememory retrieval Both trial types were presented in thestudy phase (SndashS as W1P and IndashS as W1B) and in thelie test and both were endorsed in the test phase (SndashScorrectly endorsed and IndashS incorrectly endorsed) Simi-larly a contrast between IndashS and IndashN can highlight dif-ferences between true and false memory retrieval by in-dicating differences between the false endorsement of animagined picture as being real and the correct rejectionof an imagined picture as having been only imagined (inboth of these conditions the stimuli have been treatedidentically prior to the test phase) Finally by contrast-ing true retrieval of pictures and correct rejection of onlyimagined pictures differences between SndashS and IndashN ac-tivations are suggestive of accurate retrieval However itshould be noted that although this contrast is often usedto assess memory retrieval success it can be contami-nated and weakened by activity associated with inciden-tal encoding during the retrieval task (Stark amp Okado2003)

Figure 2b shows the mean reaction times (RT) for alltrial types In the three conditions of interest the mean

RTs were fastest for SndashS items (1313 msec) and slow-est for IndashS items (1469 msec) A repeated measuresANOVA showed that there was a significant effect ofthese three trial types on RT [F(113) 5 225 p 001]Pairwise t tests showed significant RT differences betweenSndashS and IndashS [156 msec t(13) 5 2474 p 01] be-tween IndashS and IndashN [95 msec t(13) 5 271 p 05] andbetween SndashS and IndashN [61 msec t(13) 5 2306 p 01]

f MRI ResultsA voxel-wise two-factor ANOVA was first conducted

on the whole brain data to identify regions that showedany activation differences among the seven trial typesResults of this analysis identified 12 areas in which ac-tivity significantly differed across trial types (Figure 3a)These areas were treated as functionally defined ROIsand the average hemodynamic response functions foreach trial type were calculated for each region

The 12 regions identified demonstrated three distinctpatterns of activity in the functional ROI analysis withrespect to our three conditions of interest The first no-table pattern observed in the identified regions was sim-ilar increased activity for SndashS and IndashS in comparisonwith IndashN (Figure 3c) This pattern of activity was ob-served in the left parietal cortex and left frontal regionsRegions in which SndashS and IndashN differed significantly andIndashS and IndashN differed significantly included the left pari-etal cortex [BA 73940 t(13) 5 329 p 01 andt(13) 5 257 p 05 respectively] left precentral gyrus[BA 9 t(13) 5 245 p 05 and t(13) 5 319 p 01respectively] and left inferior frontal gyrus [BA 1046t(13) 5 261 p 05 and t(13) 5 298 p 05 respec-tively] The left caudate showed this similar pattern ofactivation with SndashS and IndashS demonstrating increasedactivity in comparison with IndashN but showed no signifi-

Figure 2 Behavioral data from the test phase Percentage of trials endorsed as having beenpresented with a picture at time of study (a) and reaction time for the decision (b) are plot-ted for each trial type as means across all 14 participants Error bars indicate the standarderrors of the means

fMRI OF FALSE MEMORY 329

Figure 3 (a and b) Locations and sample hemodynamic responses for regions where activity at testvaried among the seven trial types analyzed (there were too few trials in the UndashS condition to ana-lyze) (a) The 12 regions where activity varied as a function of trial type are shown as colored overlayson 3-D renderings of a brain A left parietal cortex (BA 73940) B left precentral gyrus (BA 9) Cleft inferior frontal gyrus (BA 1046) D left caudate E right middle occipital gyrus (BA 1819) Fleft cuneus (BA 171819) G left lingual gyrus (BA 19) H left fusiform gyrus (1819) I bilaterallingual gyrus (BA 18) J bilateral cuneus (BA 17182730) K right posterior parahippocampalgyrus (BA 30353637) L right anterior cingulate gyrus (BA 6932) (b) Average hemodynamic re-sponse functions (sum of beta coefficients vs image acquisition [TR]) for the three trial types of in-terest (SndashS IndashS and IndashN) are shown for three sample regions (left precentral gyrus left cuneus andright anterior cingulate) that demonstrate the patterns of activity observed (cndashe) Activity for all seventrial types analyzed in all 12 functionally defined ROIs Bars show the mean fMRI response (sum ofbeta coefficients) across participants and error bars show the standard errors of the means Aster-isks indicate significant differences in activity between conditions of interest (SndashS IndashS and IndashN) (c)Regions demonstrating the first pattern of activity included A left parietal cortex B left precentralgyrus C left inferior frontal gyrus and D left caudate (d) Regions demonstrating the second pat-tern of activity included E right middle occipital gyrus F left cuneus G left lingual gyrus H leftfusiform gyrus I bilateral lingual gyrus J bilateral cuneus and K right posterior parahippocampalgyrus (e) The third pattern of activity was observed in L right anterior cingulate gyrus

B L precentralgyrus (BA 9)

F L cuneus(BA 171819)

L R anteriorcingulate (BA 6932)

TR (25 sec)

beta

wei

ghts

b

c

d

050

ndash05ndash1

ndash15

08

04

0

05

0

ndash05

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

4

2

0

2

2

0

ndash2

ndash4

2

0

ndash2

ndash4

4

2

0

ndash2

A B C D

E F G H

I J K e L

S-S

I-S

I-N

S-S(No MI)

S-N

S-N(No MI)

U-N

1 2 3 4 5 6 7 8

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

fMRI OF FALSE MEMORY 329

Figure 3 (a and b) Locations and sample hemodynamic responses for regions where activity at testvaried among the seven trial types analyzed (there were too few trials in the UndashS condition to ana-lyze) (a) The 12 regions where activity varied as a function of trial type are shown as colored overlayson 3-D renderings of a brain A left parietal cortex (BA 73940) B left precentral gyrus (BA 9) Cleft inferior frontal gyrus (BA 1046) D left caudate E right middle occipital gyrus (BA 1819) Fleft cuneus (BA 171819) G left lingual gyrus (BA 19) H left fusiform gyrus (1819) I bilaterallingual gyrus (BA 18) J bilateral cuneus (BA 17182730) K right posterior parahippocampalgyrus (BA 30353637) L right anterior cingulate gyrus (BA 6932) (b) Average hemodynamic re-sponse functions (sum of beta coefficients vs image acquisition [TR]) for the three trial types of in-terest (SndashS IndashS and IndashN) are shown for three sample regions (left precentral gyrus left cuneus andright anterior cingulate) that demonstrate the patterns of activity observed (cndashe) Activity for all seventrial types analyzed in all 12 functionally defined ROIs Bars show the mean fMRI response (sum ofbeta coefficients) across participants and error bars show the standard errors of the means Aster-isks indicate significant differences in activity between conditions of interest (SndashS IndashS and IndashN) (c)Regions demonstrating the first pattern of activity included A left parietal cortex B left precentralgyrus C left inferior frontal gyrus and D left caudate (d) Regions demonstrating the second pat-tern of activity included E right middle occipital gyrus F left cuneus G left lingual gyrus H leftfusiform gyrus I bilateral lingual gyrus J bilateral cuneus and K right posterior parahippocampalgyrus (e) The third pattern of activity was observed in L right anterior cingulate gyrus

B L precentralgyrus (BA 9)

F L cuneus(BA 171819)

L R anteriorcingulate (BA 6932)

TR (25 sec)

beta

wei

ghts

b

c

d

050

ndash05ndash1

ndash15

08

04

0

05

0

ndash05

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

4

2

0

2

2

0

ndash2

ndash4

2

0

ndash2

ndash4

4

2

0

ndash2

A B C D

E F G H

I J K e L

S-S

I-S

I-N

S-S(No MI)

S-N

S-N(No MI)

U-N

1 2 3 4 5 6 7 8

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

330 OKADO AND STARK

cant differences between SndashS and IndashS between IndashS andIndashN or between SndashS and IndashN

The second pattern of activity (Figure 3d) was associ-ated with similar levels of activity for SndashS items and IndashNitems that was substantially greater than the activity as-sociated with IndashS items This pattern of activity was ob-served primarily in the occipital region as well as in aportion of the parahippocampal gyrus just posterior tothe parahippocampal cortex Regions in which SndashS andIndashS differed significantly and IndashS and IndashN differed signif-icantly included right middle occipital gyrus [BA 1819t(13) 5 333 p 01 and t(13) 5 2352 p 01 re-spectively] left cuneus [BA171819 t(13) 5 353 p 01 and t(13) 5 2347 p 01 respectively] left lin-gual gyrus [BA 19 t(13) 5 448 p 01 and t(13) 52499 p 001 respectively] left fusiform gyrus[BA 1819 t(13) 5 257 p 05 and t(13) 5 2291p 05] and right posterior parahippocampal gyrus[BA 30353637 t(13) 5 444 p 01 and t(13) 52457 p 01 respectively] Bilateral lingual gyrus(BA 18) showed significant activation differences foronly the IndashS and IndashN [t(13) 5 2318 p 01] contrastBilateral cuneus (BA 17182330) elicited the same ac-tivation pattern but showed no significant differencesbetween SndashS and IndashS between IndashS and IndashN or betweenSndashS and IndashN

The third pattern of activity noted in the identified re-gions was increased activity for IndashS items in comparisonwith SndashS and IndashN items (Figure 3e) This pattern of ac-tivity was observed in the right anterior cingulate gyrus(BA 6932) This region showed significant activationdifferences between SndashS and IndashS [t(13) 5 2548 p 001] and between IndashS and IndashN [t(13) 5 325 p 01]

A separate analysis was done on the data from theMTL since the MTL plays a vital role in declarativememory tasks such as those used in the current experi-ment (see Milner et al 1998 for a review) Using theROIndashAL technique (Stark amp Okado 2003) to optimizethe alignment and statistical power within the MTL aseparate voxel-wise two-factor ANOVA was performedon the aligned data with a clustering threshold adjustedto reflect the number of voxel-wise comparisons withinthe MTL Four regions in the MTL showed activity dif-ferences across the trial types Two regions showed ac-tivity similar to that of the second activation pattern dis-cussed That is SndashS and IndashN items were similarly moreactive in comparison with IndashS items A region within theleft temporopolar portion of perirhinal cortex showedsignificant activation differences between SndashS (betaweight 5 074 SEM 5 023) and IndashS [beta weight 5217472 SEM 5 041 t(13) 5 272 p 05] and be-tween IndashS and IndashN [beta weight 5 2029 SEM 5 031t(13) 5 233 p 01] Right parahippocampal gyrus(also observed in the whole-brain ANOVA results) showedsignificant activation differences between SndashS (betaweight 5 2027 SEM 5 047) and IndashS [beta weight 52248 SEM 5 0592 t(13) 5 559 p 001] and be-tween IndashS and IndashN [beta weight 5 2064 SEM 5 047

t(13) 5 2618 p 001] This region included a portionof the most posterior extent of the parahippocampal cor-tex and extended significantly posterior to our definitionof the parahippocampal cortex with the majority of theregion lying outside of the parahippocampal cortex Thetwo other regions identified in the MTL an area withinthe right temporopolar portion of perirhinal cortex andan area within left entorhinal cortex did not show anysignificant activation differences across any of the trialtypes of interest For both of these regions however ac-tivity for SndashN differed significantly from that of the restof the trial types [specifically SndashN (no MI) for righttemporopolar cortex was least active and SndashN for leftentorhinal cortex was most active]

DISCUSSION

In this study we compared the neural bases of memoryretrieval processes that yielded true and false memoriesThree key contrasts were examined (1) neural differencesbetween true recognition of pictures that had been seenand false recognition of pictures that had been onlyimagined (SndashS vs IndashS respectively) (2) false recogni-tion of imagined pictures and correct rejection of imag-ined pictures (IndashS vs IndashN respectively) and (3) truerecognition of seen pictures and correct rejection ofimagined pictures (SndashS vs IndashN respectively) The firsttwo contrasts help identify differences between true andfalse memory retrieval whereas the third helps identifyactivity associated with accurate memory retrieval

We observed that certain regions of the brain showedstrikingly distinct patterns of activation for retrieval pro-cesses that yield true and false memories In left parietaland left frontal cortices SndashS and IndashS trials were simi-larly more active than IndashN trials Upon further investi-gation left parietal and left frontal activity differed out-side of our conditions of interest in that the left parietalcortex showed low levels of activity associated withnovel UndashN items Bilateral occipital regions and rightposterior parahippocampal gyrus (posterior to para-hippocampal cortex) showed significantly increased ac-tivity for both SndashS and IndashN trials in comparison with IndashStrials In right anterior cingulate gyrus there was moreactivity for IndashS trials than for SndashS and IndashN trials Thesedistinct neural response patterns are suggestive of dif-ferential processing by the various brain regions for thesame information

Before we turn to a discussion of the possible sourcesof activity for SndashS IndashS and IndashN trials it will be usefulto understand the potential effects of the misinformationtask used in the lie test This misinformation phase wasdeliberately employed to convert IndashN trials into IndashS tri-als (and potentially SndashN trials into SndashS trials) and itwas successful in increasing the overall false memoryrate It is clear that from the data at hand it would be dif-ficult if not impossible to understand what cognitive orneural processes were engaged during this task and howthese processes have affected the results However since

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

fMRI OF FALSE MEMORY 331

the primary interest of this study was to examine thefalse memory reconstructive retrieval process and nothow encoding affects false memories the use of the lietest should not affect what one can conclude about ac-tivity for IndashS items during the test phase We would cau-tion careful interpretation wherever a difference betweenSndashS and SndashS (no MI) activity was observed Althoughboth represent accurate memory retrieval SndashS (no MI)trials were not presented during the lie test whereas SndashStrials were If activity differs for SndashS and SndashS (no MI) tri-als we must clearly consider the influence of the misin-formation task on brain activity

In addition to any effects of the misinformation taskit is worthwhile to note that activity from our trials of in-terest could be due to several possible sources SndashS ac-tivity (eg episodic memory of picture during studyphase or vivid memory of picture without episodic com-ponent) IndashS activity (eg false episodic memory of pic-ture during study phase or vivid memory of picture with-out episodic component) and IndashN activity (eg episodicmemory of no picture during study phase or poor imageof picture retrieved)

With these characterizations in mind we can attemptto understand the computational implications of thethree observed patterns of activity The first activationpattern observed for the three conditions of interest (Fig-ure 3c) is consistent with the hypothesis that activity inleft parietal and left frontal regions is affected by theamount of information retrieved (episodicsource anditem-based components of the memory) According tothis hypothesis one would predict that not only shouldSndashS and IndashS activity be high relative to IndashN but UndashN ac-tivity should also be lowest The left parietal cortexdemonstrated this pattern of a graded response of howmuch contextual information was recovered or believedto be recovered When the participants believed they hadstudied a picture regardless of whether they in fact hadthere was a trend in this region for activity to be great-est Items that were studied but forgotten and those thatwere correctly rejected showed a trend of less activitysuggesting that the participants remembered studyingthe words but did not remember them in enough detail tobelieve they had seen pictures Novel items that werecorrectly rejected showed the least amount of activitysuggesting that the participants could recall little or nocontextual information associated with the novel wordsSimilar results have been obtained by others investigat-ing true and false memories and imagery but those au-thors reported that retrieval activity that resulted in truememories was greater than activity that resulted in falsememories (Allan et al 1998 Cabeza et al 2001 Gon-salves amp Paller 2000 Henson et al 1999 Nessler et al2001) It may be that the participants in the present studyrecalled as much perceptual sensory and contextual de-tail about the imagined objects as they did about the per-ceived objects Wheeler and Buckner (2003) found that aslong as participants perceived an item as old left parietalcortex was active regardless of accuracy modality of item

presentation and number of times the item was studiedAlthough Johnson and Raye (1981) suggest that per-ceived and imagined events differ in a number of ways(eg quality and quantity of detail recovered cognitiveoperations used to form memory) it appears that activ-ity of the left parietal cortex was correlated with theamount of information recovered or believed to be recov-ered from memory

In the left frontal regions however SndashS IndashS and IndashNactivity showed the same pattern whereas UndashN activitywas not significantly lower than IndashN activity These re-gions may instead be associated with response monitor-ing of information believed to have been studied or suc-cessful source retrieval If so SndashS and IndashS activityshould be similar and higher than all other conditionssince in each of these trial types participants are retriev-ing pictures and images believed to have been seen Thispattern was indeed found in the left frontal regions andis consistent with Cabeza et alrsquos (2001) observation thattrue and false recognition evoked activity in bilateraldorsolateral prefrontal cortex Similarly other studieshave reported activity in left anterior prefrontal cortexthat is associated with the retrieval of perceptual infor-mation (Ranganath Johnson amp DrsquoEsposito 2000) Theprefrontal cortex is also frequently associated with suc-cessful memory retrieval including accurate episodic re-membering (Buckner Koutstaal Schacter Dale et al1998 Buckner Koutstaal Schacter Wagner amp Rosen1998 Cabeza amp Nyberg 2000 Ranganath et al 2000)Thus the activity observed in the frontal regions in thepresent study is consistent with several findings in theliterature suggesting that these regions are associatedwith response monitoring or successful source retrievalAs is suggested by Johnsonrsquos (Johnson Hashtroudi ampLindsay 1993 Johnson amp Raye 2000) SMF and Schac-terrsquos (Schacter et al 1998) CMF recovered informationmust undergo an evaluation and criterion-setting processto determine the accuracy of the memory

The second activation pattern observed primarily invisual areas (Figure 3d) although quite robust is diffi-cult to interpret from the present data The pattern of thethree conditions of interest suggests that numerous oc-cipital regions and the posterior right parahippocampalgyrus contain information about the difference betweentrue and false responses since IndashS trials showed signif-icantly less activity than SndashS or IndashN trials (except in bi-lateral lingual gyrus and bilateral cuneus where thesame trend was observed but the differences were notsignificant) By facilitating an examination of the pre-dictions a hypothesis such as this would make for theother four stimulus conditions the present design allowsus to test such a hypothesis to a greater extent than haspreviously been possible in the neuroimaging of falsememory For example if activity were simply a functionof the truthfulness of the response one might predict thatall accurate responses should be equally active andgreater in activity than all inaccurate responses whichalso should be equally active It is also possible that re-

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

332 OKADO AND STARK

trieval of seen pictures elicited more activity than re-trieval of imagined pictures believed to have been seenPerceived objects are rich in sensory detail whereas imag-ined objects tend to lack such perceptual detail (Johnsonamp Raye 1981) therefore it is not surprising that truememories evoked more activity in the occipital regionthan did false memories of imagined pictures In such re-ality monitoring studies it is often found that retrieval ofreal pictures contains more details and information thanretrieval of imagined pictures (Johnson amp Raye 1981)Furthermore although the retrieval of imagined pictureshas elicited activity in both primary and higher order vi-sual cortex (Kosslyn et al 1999 Wheeler et al 2000)some reports on whether imagined pictures elicit primaryvisual cortical activity have been inconsistent with eachother (see Cabeza amp Nyberg 2000 for a review) raisingthe possibility that imagined pictures do not possess allthe elements of perceived pictures necessary for thebrain to respond as if a real picture were being remem-bered It is also possible that the amount of activity is de-termined by the total amount of study exposure It is cer-tainly possible that one effect of retrieving and reencodingthese stimuli in the lie test was to enhance the strengthand vividness of the memory for these pictures Simi-larly the increase in the false memory rate resulting fromthe lie test is consistent with the idea that the misinfor-mation phase enhanced the vividness of the participantsrsquomemory for items initially only imagined The differencebetween SndashS and SndashS (no MI) activity in all the areas inthe occipital lobes (except left fusiform gyrus) and rightposterior parahippocampal gyrus is indicative of a sub-stantial role of the lie test be it in the form of a second exposure or as a more complex effect It is difficult tointerpret any influence on neural activity that the misin-formation effect may have on these brain regions butthis activation pattern was robust and prevalent sug-gesting that these regions are processing the retrieval ofreal and imagined pictures in a systematic fashion Thepresence of the additional five-stimulus conditions inour experimental design (four of which can be analyzed)has allowed us to reject all three of these hypotheseswhen they are considered in isolation Thus although thispattern was clearly robust it appears to be the result ofan interaction of the factors discussed above (or otherunknown factors) that is beyond explanation from the pres-ent data Further experimentation designed to more di-rectly address this issue is therefore warranted

The third observed pattern of activity of the three con-ditions of interest (Figure 3e) in the right anterior cingu-late gyrus suggests that this region is heavily associatedwith effort The neural activation and behavioral RT pat-tern demonstrate a strong correlation in that IndashS activityand RT are highest and SndashS activity and RT are lowestand the other trial types fall in between (Figure 2b andFigure 3e) Interestingly this region has been identifiedto be active during erroneous responses as well as duringcorrect responses when there is a high level of responsecompetition or conflict (Botvinick Nystrom Fissell

Carter amp Cohen 1999 Carter et al 1998) For exampleCarter et al demonstrated that the anterior cingulate cor-tex detects situations in which errors are likely to occurA prediction of this hypothesis is that SndashS and UndashN tri-als should elicit the smallest amount of activation fol-lowed by SndashS (no MI) and IndashN trials followed by activ-ity associated with SndashN trials and finally by IndashS trials(which were associated with the longest RTs are erro-neous responses and have clear potential for competi-tion between conflicting responses) Exactly this patternwas observed in the data This interpretation of effortand high conflict driving the right anterior cingulate isclearly consistent with the data since this region was theonly one to demonstrate increased activity for false mem-ories in comparison with the other trial types It is rea-sonable to assume that high levels of conflict and effortare involved when imagined pictures are believed to bereal

It is worth noting that although the MTL is heavily as-sociated with memory encoding and retrieval processes(see eg Milner et al 1998) we did not observe promi-nent activity throughout the MTL as one might expect ina declarative memory retrieval task such as the one usedhere It is important to note that the literature on mem-ory retrieval is marked by numerous such failures to ob-serve MTL activity since the contrasts used to index re-trieval success are often confounded with incidentalencoding processes (Stark amp Okado 2003 Stark ampSquire 2000 2001a) Here the lack of hippocampal andother MTL activity may ultimately be due to confound-ing effects such as incidental encoding or the presence ofepisodic or source memory components in all the trialtypes analyzed Such a commonality may have resultedin similar levels of activity across the trial conditions inmany MTL regions Alternatively it is quite possible thattrue and false recognition are not differentiated by theMTL If the role of the MTL is to integrate pieces of in-formation and later help recover these pieces via this re-constructive retrieval process a misattributed memorymay be indistinguishable at this level

We should note that although large regions of theMTL were not active differentially across trial types itwas not the case that we observed no differential activ-ity in the MTL at all In an analysis optimized for the MTLwe did observe MTL activity that varied by trial type inposterior right parahippocampal gyrus right entorhinalcortex and bilateral temporopolar cortex We shouldnote that although the activity in the right parahippo-campal gyrus extended to include some of the right para-hippocampal cortex it was largely posterior to what wedefine as parahippocampal cortex Nonetheless the pat-tern of activity in right parahippocampal gyrus and lefttemporopolar cortex was strikingly similar to the patternobserved in the occipital regions (Figure 3d) Such para-hippocampal gyrus activations (often it is unclearwhether the findings lie within the parahippocampal cor-tex) have been reported to occur during memory retrieval(see eg Cabeza et al 2001 Schacter et al 1997

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

fMRI OF FALSE MEMORY 333

Stark amp Okado 2003) Cabeza et al (2001) suggestedthat this region responds to sensory details of the infor-mation retrieved and can therefore differentiate betweenmemories of a perceived stimulus and memories of aninternally generated stimulus because a presented stim-ulus carries more sensory details than one that was notpresented which should at best seem familiar In ac-cordance with Cabeza et alrsquos results the parahippo-campal gyrus activity from this study did show more ac-tivity for SndashS than for IndashS trials However it also showedmore activity for SndashS than for SndashS (no MI) trials andequal activity for SndashS SndashN and UndashN trials which doesnot concur with the explanation offered by Cabeza et alAlthough the tasks (DRM vs reality monitoring) clearlydiffered it is unclear how this could have affected the re-sults given the hypothesis proposed by Cabeza et alsince the present task seems to provide ample opportu-nity to assess activity for externally perceived versus in-ternally generated stimuli as well The pattern of activ-ity observed here (and potentially those reported byCabeza et al given the marked resemblance to the pat-tern of activity observed in the occipital region) suggeststhat the MTL activations observed in this study may notbe the result of MTL memory processes but rather theresult of visual cortexrsquos feeding information into theseregions (Suzuki amp Eichenbaum 2000) Logically theconversemdashthat is that these MTL regions are driving theactivity observed in visual cortexmdashcould also be trueWith its poor temporal resolution f MRI cannot clearlydifferentiate between the two

This study has demonstrated that occipital and rightposterior parahippocampal gyrus showed greater activityfor seen pictures than for imagined pictures believed tobe seen In contrast the right anterior cingulate regionsshowed the opposite pattern These data suggest a distinc-tion between retrieval processes that yield true and falsememories Finally the left parietal lobe left frontal lobeand MTL looked similar for seen pictures and for imag-ined pictures believed to be seen suggesting an inabilityto detect differences between retrieval processes thatyield true memories and those that yield false memories

Thus the data suggest that for true and false memoriesof pictures the occipital and posterior parahippocampalgyrus regions show activity that distinguishes thesememories whereas the left frontal and parietal activityserve as an index of how much both true and false mem-ories are believed to be true The anterior cingulate gyrusshows activity that isolates memories that are associatedwith effort and conflict which tend to be the false mem-ories in this type of reality monitoring paradigm Howthese various regions work together during retrieval toyield true and false memories is a topic for future research

REFERENCES

Allan K Wilding E L amp Rugg M D (1998) Electrophysiolog-ical evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Botvinick M Nystrom L E Fissell K Carter C S amp Cohen

J D (1999) Conflict monitoring versus selection-for-action in ante-rior cingulate cortex Nature 402 179-181

Buckner R L Koutstaal W Schacter D L Dale A MRotte M amp Rosen B R (1998) Functional-anatomic study ofepisodic retrieval II Selective averaging of event-related fMRI tri-als to test the retrieval success hypothesis NeuroImage 7 163-175

Buckner R L Koutstaal W Schacter D L Wagner A D ampRosen B R (1998) Functional-anatomic study of episodic retrievalusing fMRI I Retrieval effort versus retrieval success NeuroImage7 151-162

Cabeza R amp Nyberg L (2000) Imaging cognition II An empiricalreview of 275 PET and fMRI studies Journal of Cognitive Neuro-science 12 1-47

Cabeza R Rao S M Wagner A D Mayer A R amp SchacterD L (2001) Can medial temporal lobe regions distinguish true fromfalse An event-related functional MRI study of veridical and illu-sory recognition memory Proceedings of the National Academy ofSciences 98 4805-4810

Carter C S Braver T S Barch DM Botvinick MM NollDamp Cohen J D (1998) Anterior cingulate cortex error detectionand the online monitoring of performance Science 280 747-749

Cox R W (1996) AFNI Software for analysis and visualization offunctional magnetic resonance neuroimages Computational Bio-chemical Research 29 162-173 Available at httpafninimhnihgov

Deese J (1959) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22

Fabiani M Stadler M A amp Wessels P M (2000) True but notfalse memories produce a sensory signature in human lateralizedbrain potentials Journal of Cognitive Neuroscience 12 941-949

Farah M J (1989) The neural basis of mental imagery Trends inNeurosciences 12 395-399

Gonsalves B amp P aller K A (2000) Neural events that underlie re-membering something that never happened Nature Neuroscience 31316-1321

Henson R N A Rugg M D Shallice T Josephs O amp DolanR J (1999) Recollection and familiarity in recognition memory Anevent-related functional magnetic resonance imaging study Journalof Neuroscience 19 3962-3972

Insausti R Juottonen K Soininen H Insausti A M P arta-nen K Vainio P Laakso M P amp P itkanen A (1998) MR vol-umetric analysis of the human entorhinal perirhinal and temporo-polar cortices American Journal of Neuroradiology 19 659-671

Johnson M K (1997) Source monitoring and memory distortionPhilosophical Transactions of the Royal Society of London Series B352 1733-1745

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85

Johnson M K amp Raye C L (2000) Cognitive and brain mecha-nisms of false memories and beliefs In D L Schacter amp E Scarry(Eds) Memory and belief (pp 36-86) Cambridge MA HarvardUniversity Press

Johnson M K Raye C L Wang A Y amp Taylor T H (1979)Fact and fantasy The roles of accuracy and variability in confusingimaginations with perceptual experiences Journal of ExperimentalPsychology Human Learning amp Memory 5 229-240

Johnson MK Taylor T H amp Raye C L (1977) Fact and fantasyThe effects of internally generated events on the apparent frequencyof externally generated events Memory amp Cognition 5 116-122

Kosslyn S M P ascual-Leone A Felician O Camposano SKeenan J P Thompson W L Ganis G Sukel K E amp AlpertN M (1999) The role of Area 17 in visual imagery Convergent ev-idence from PET and rTMS Science 284 167-170

Loftus E F amp Loftus G R (1980) On the permanence of stored information in the human brain American Psychologist 35 409-420

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

334 OKADO AND STARK

Loftus E F amp P almer J C (1974) Reconstruction of automobiledestruction An example of the interaction between language andmemory Journal of Verbal Learning amp Verbal Behavior 13 585-589

Loftus E F amp P ickrell J E (1995) The formation of false mem-ories Psychiatric Annals 25 720-725

McCloskey M amp Zaragoza M (1985) Misleading postevent in-formation and memory for events Arguments and evidence againstmemory impairment hypotheses Journal of Experimental Psychol-ogy General 114 1-16

Miller A R Baratta C Wynveen C amp Rosenfeld J P (2001)P300 latency but not amplitude or topography distinguishes betweentrue and false recognition Journal of Experimental PsychologyLearning Memory amp Cognition 27 354-361

Milner B Squire L R amp Kandel E R (1998) Cognitive neuro-science and the study of memory Neuron 20 445-468

Nessler D Mecklinger A amp P enney T B (2001) Event relatedbrain potentials and illusory memories The effects of differential en-coding Cognitive Brain Research 10 283-301

P ruessner J C Kohler S Crane J P ruessner M Lord CByrneA Kabani N Collins D L amp Evans A C (2002) Vol-umetry of temporopolar perirhinal entorhinal and parahippocampalcortex from high-resolution MR images Considering the variabilityof the collateral sulcus Cerebral Cortex 12 1342-1353

Ranganath C Johnson M K amp Drsquo Esposito M (2000) Left an-terior prefrontal activation increases with demands to recall specificperceptual information Journal of Neuroscience 20 1-5

Roediger H L III (1996) Memory illusions Journal of Memory ampLanguage 35 76-100

Roediger H L III amp McDermott K B (1995) Creating falsememories Remembering words not presented in lists Journal of Ex-perimental Psychology Learning Memory amp Cognition 21 803-814

Rugg M D Fletcher P C Frith C D Frackowiak R S J ampDolan R J (1996) Differential activation of the prefrontal cortexin successful and unsuccessful memory retrieval Brain 1919 2073-2083

Schacter D L (1999) The seven sins of memory American Psy-chologist 54 182-203

Schacter D L Buckner R L Koutstaal W Dale A M ampRosen B R (1997) Late onset of anterior prefrontal activity duringtrue and false recognition An event-related fMRI study NeuroImage6 259-269

Schacter D L Norman K A amp Koutstaal W (1998) The cog-

nitive neuroscience of constructive memory Annual Review of Psy-chology 49 289-318

Schacter D L Reiman E Curran T Yun L S Bandy D Mc-Dermott K B amp Roediger H L III (1996) Neuroanatomicalcorrelates of veridical and illusory recognition memory Evidencefrom positron emission tomography Neuron 17 267-274

Stark C E L amp Okado Y (2003) Making memories without tryingMedial temporal lobe activity associated with incidental memory for-mation during recognition Journal of Neuroscience 23 6748-6753

Stark C E L amp Squire L R (2000) Functional magnetic reso-nance imaging (fMRI) activity in the hippocampal region duringrecognition memory Journal of Neuroscience 20 7776-7781

Stark C E L amp Squire L R (2001a) Simple and associative recog-nition memory in the hippocampal region Learning amp Memory 8190-197

Stark C E L amp Squire L R (2001b) When zero is not zero Theproblem of ambiguous baseline conditions in fMRI Proceedings ofthe National Academy of Sciences 98 12760-12766

Suzuki W A amp Eichenbaum H (2000) The neurophysiology ofmemory Annals of the New York Academy of Sciences 911 175-191

Talairach J amp Tournoux P (1988) Co-planar stereotaxic atlas ofthe human brain New York Thieme

Tulving E (1985) Memory and consciousness Canadian Psychol-ogy 26 1-12

Wagner A D Desmond J E Glover G H amp Gabrieli J D E(1998) Prefrontal cortex and recognition memory Functional-MRIevidence for context-dependent retrieval processes Brain 121 1985-2002

Ward D (2000) Deconvolution analysis of FMRI time series dataAvailable at httpafninimhnihgovafnidocpdf3dDeconvolvepdf)

Wheeler M E amp Buckner R L (2003) Functional dissociationamong components of remembering Control perceived oldness andcontent Journal of Neuroscience 23 3869-3880

Wheeler M E P etersen S E amp Buckner R L (2000) Memo-ryrsquos echo Vivid remembering reactivates sensory-specific cortexProceedings of the National Academy of Sciences 97 11125-11129

(Manuscript received June 19 2003revision accepted for publication December 12 2003)