training of somatosensory discrimination after strokeof somatosensory deﬁcits, particularly touch...

Authors:Leeanne M. Carey, PhDThomas A. Matyas, PhD

Affiliations:From the National Stroke ResearchInstitute, Heidelberg West, Victoria,Australia (LMC); and the Schools ofOccupational Therapy (LMC) andPsychological Sciences (TAM),LaTrobe University, Bundoora,Victoria, Australia.

Disclosures:Supported, in part, by a LaTrobeUniversity Faculty of Health SciencesResearch Grant, an AustralianResearch Council small grant, aResearch Development Grant fromthe Department of Health andHuman Services, and a NationalHealth and Medical Research Councilof Australia project grant (191214).

Presented, in part, at the 11thInternational Congress of the WorldFederation of OccupationalTherapists, London, UK, and at theThird Annual Perception for ActionConference, Melbourne, Australia.

Correspondence:All correspondence and requests forreprints should be addressed toLeeanne M. Carey, PhD, NationalStroke Research Institute, Level 2,Neurosciences Building, AustinHealth, Banksia Street, HeidelbergWest, Victoria, Australia 3081.

0894-9115/05/8405-0001/0American Journal of PhysicalMedicine & RehabilitationCopyright © 2005 by LippincottWilliams & Wilkins

DOI: 10.1097/01.PHM.0000159971.12096.7F

Training of SomatosensoryDiscrimination After StrokeFacilitation of Stimulus Generalization

ABSTRACT

Carey LM, Matyas TA: Training of somatosensory discrimination after stroke: Facil-itation of stimulus generalization. Am J Phys Med Rehabil 2005;84:000–000.

Objective: Task-specific learning typifies perceptual training but limits rehabilitation ofsensory deficit after stroke. We therefore investigated spontaneous and procedurallyfacilitated transfer of training effects within the somatosensory domain after stroke.

Design: Ten single-case, multiple-baseline experiments were conducted withstroke participants who had impaired discrimination of touch or limb-positionsense. Each experiment comprised three phases: baseline, stimulus-specifictraining of the primary discrimination stimulus, and either stimulus-specific train-ing of the transfer stimulus or stimulus-generalization training. Both the trainedand transfer stimuli were monitored throughout using quantitative, norm-refer-enced measures. Data were analyzed using individual time-series analysis andmeta-analysis of intervention effects across case experiments.

Results: Stimulus-specific training was successful for trained texture and proprio-ceptive discriminations, but it failed to show spontaneous transfer to related untrainedstimuli in the same modality in seven of eight experiments in which this was possible.In contrast, intramodality transfer was obtained with stimulus-generalization training infour of five experiments that investigated stimulus-generalization training of texturediscrimination. Findings were confirmed by meta-analysis.

Conclusions: Our findings demonstrate generalization of training within a somato-sensory modality poststroke, provided that a program designed to enhance transferis used. This has implications for the design of efficient rehabilitation programs.

Key Words: Sensation, Rehabilitation, Cerebrovascular Accident, Transfer of Training

balt5/z79-phm/z79-phm/z7900505/z792796-05a browncm S�10 3/11/05 4:48 Art: 162531

May 2005 Training of Somatosensory Discrimination 1

RESEARCH ARTICLE

Stroke

<ARTICLE DOCTOPIC��Research Articles� DOCSUBJ��Stroke� DATE��May 2005� VID��84� ISS��5� PPF��●●●� PPL��●●●�DOI��10.1097/01.PHM.0000159971.12096.7F�>

Generalization of intervention effects to un-trained tasks is central to sensorimotor neuroreha-bilitation. Task-specific training, although effec-tive, has the potential to be very costly, and noveltasks will present continuing problems. Discoveryof effective training methods also able to achievetransfer of gains to novel tasks is thus essential.Furthermore, investigation of the boundaries oftransfer is increasingly regarded as a key ingredientto a proper understanding of learning effects in thesensorimotor system.1 In the sensory domain, in-vestigation of stimulus-generalization effects couldhelp clarify what is encoded about the stimuli dur-ing the training experience.

“Generalization of training” or “transfer oflearning” to an unpracticed task has been investi-gated with healthy subjects in various domains,including perception.2 Empirical investigations inthe perceptual learning literature suggest thattransfer is possible in some instances.3,4 However,discrimination training is often highly specific tothe task.2,3,5 A general finding is that similaritybetween the trained and transfer tasks is an impor-tant determinant of positive transfer.2

The dimensions of similarity that seem to mat-ter include whether the novel transfer task iswithin the same perceptual dimension as thetrained task, the similarity of the required re-sponse, and the similarity in method of processingthe information.2–4 Body location, density of recep-tors, and organization of the somatosensory cortexmay also affect transfer of learning in somatosen-sory discrimination.2–4 The training conditionsunder which transfer might be expected also needto be considered. Principles that may enhancetransfer of training have been identified in themotor-learning literature.1 However, exposure andfeedback may be adequate for perceptual transfer,at least for unimpaired subjects.3–6

Generalization of training within the sameperceptual dimension or sensory modality has re-ceived only limited investigation in stroke patients.Spontaneous improvement in related visual func-tions was found in a controlled study that trained asingle visual function of patients with cerebralblindness.7 In contrast, we found stimulus-specifictraining (SST) effects across tactile and propriocep-tive discrimination tasks poststroke.8 Studies thathave trained multiple modalities simultaneously,used objects with multisensory demands,9 or em-ployed sensorimotor tasks10,11 have reported im-provement in untrained sensory tasks, suggestingthe potential for generalized somatosensory effects.These results are consistent with findings of spon-taneous transfer across multidimensional tasks inhealthy subjects.2 Although spontaneous transfermay be achieved by unimpaired subjects when

tasks and processing activities are highly similar, itis unknown whether these conditions are adequatefor individuals in whom the perceptual system isdamaged. Furthermore, the potential for transferof training under different training conditions hasnot been systematically investigated within the so-matosensory domain after stroke.

We have focused our research on rehabilitationof somatosensory deficits, particularly touch andproprioceptive discrimination. These deficits arecharacteristic of the loss experienced and are re-ported in approximately 50% of stroke patients.12

Such deficits pose substantial difficulties in recep-tion of sensory information and exploration of theenvironment and have detrimental effects on spon-taneous use of hands, object manipulation, andprecision grip. Moreover, sensory loss has a nega-tive effect on personal safety, functional outcome,and quality of life, and it influences length of stayin the hospital (see Carey12 for review), highlight-ing the functional importance of training thesecapacities.

This investigation of generalization effects fol-lowed our initial findings that stroke-induced im-pairments of texture discrimination and limb-posi-tion senses are trainable.8 Improvements werespecific to the stimuli trained2 and confirmed theexpected independence of tactile and propriocep-tive performance.8 Although lack of transfer acrosstactile and proprioceptive discrimination was ex-pected, those data do not address transfer acrossstimuli of the same type, such as textures withdifferent distinctive features of roughness. Efficientretraining of somatosensory discrimination abili-ties requires an understanding of the similarityrequired to achieve successful transfer.

We therefore conducted a series of ten con-trolled, multiple-baseline, single-case experi-ments13 to investigate transfer across stimuliwithin the same sensory-perceptual dimension us-ing either SST or stimulus-generalization training(SGT). The design permitted training the primarydiscrimination stimuli while simultaneously mon-itoring the influence of this treatment on thetransfer stimuli, and the design delayed introduc-tion of generalization-enhanced training to afterthis control phase. Study 1 investigated spontane-ous transfer of training effects within tactile andproprioceptive domains after SST. Study 2 investi-gated transfer to novel stimuli within the tactiledimension using a program modified to facilitateSGT.

Transfer tasks were selected to require thesame sensory-perceptual dimension and body loca-tion as the trained task, as these features of simi-larity seem to be important in perceptual learn-ing.2,3 Tactile stimuli were two types of texturedsurfaces: plastic grids and fabrics. These stimuli


2 Carey and Matyas Am. J. Phys. Med. Rehabil. ● Vol. 84, No. 5

require processing within the same tactile domainand are both explored with the fingertip but havedifferent surface characteristics. The propriocep-tion tasks involved discrimination of imposed wristpositions in the flexion-extension and ulnar-radialdeviation planes of movement. They required thesame type of limb-position discrimination and thesame body location (i.e., the wrist), but they usedpositions within different planes of movement.Norm-referenced, quantitative, and reliable mea-sures of these tactile and proprioceptive discrimi-nations were available as outcome measures.12,14,15

MATERIALS AND METHODSStudy 1: Spontaneous Transfer ofTraining Effects with SSTSubjects

Five stroke patients with tactile or propriocep-tive discrimination impairment, as tested by themeasures described below, were studied. They weremedically stable, had adequate comprehension ofinstructions for assessment, had no peripheralneuropathy or previous central nervous systemdysfunction, and were free of unilateral spatial ne-glect based on clinical observation and standardneuropsychological tests. Patients meeting thesecriteria were selected sequentially, as they pre-sented, and gave voluntary informed consent. Theproject was approved by the human ethics commit-tees of LaTrobe University and participating hospi-tals and conformed to the Helsinki Declaration.

MaterialsMeasures of touch discrimination were the

Tactile Discrimination Test (TDT)15 and the FabricMatching Test (FMT).12 The TDT employed finelygraded plastic surfaces marked by ridges at setspatial intervals. Texture grids were presented insets of three, with two surfaces identical and onedifferent. The FMT comprised two identical sets often cotton-based fabrics, ranked from smooth torough by unimpaired subjects. These measureshave high retest reliability and good discriminativevalidity and normative standards.15,16

For the proprioception domain, the Wrist Po-sition Sense Test (WPST) quantified the subject’sability to indicate wrist position after an imposedmovement in the flexion-extension14 or ulnar-ra-dial deviation12 plane of movement. A box-like ap-paratus included two protractor scales to indicatetarget and response positions and splints for fore-arm and hand. The hand splint was attached to alever allowing freedom of movement at the wrist.The box occluded vision of subject’s wrist positionand of the examiner’s lever manipulations. Eachtest comprised 20 predetermined wrist positions.High retest reliability and good discriminative va-

lidity and normative standards have also been es-tablished for these tests.14,16

ProcedureFive single-case experiments were conducted

to investigate spontaneous transfer of training torelated stimuli in study 1. Two experiments inves-tigated transfer within the tactile-discriminationdimension and three within proprioception. In thetactile experiments, texture grids were employed asthe primary stimuli to be trained and the fabrics asthe transfer stimuli. Imposed wrist positionswithin the flexion-extension plane were the pri-mary trained proprioception stimuli and positionswithin the ulnar-radial deviation plane the transferstimuli.

The typical case experiment had three phases,each phase comprising ten sessions in which boththe trained and transfer tasks were assessed. As-sessment sessions were scheduled 48–72 hrs apart.In the first phase, both tasks were monitored only.In the second phase, the texture grid or flexion-extension stimuli were trained over ten treatmentsessions, interspersed between assessment ses-sions, while baseline monitoring continued on thetransfer response. In the third phase, treatmentwas introduced for the transfer stimuli over tensessions. This permitted investigation of the sub-ject’s potential for a SST effect on the transfer task.Follow-up was conducted after an interval equiva-lent to the combined time of baseline and inter-vention phases (i.e., 12–14 wks after the end oftraining).

Test Procedure and ScoringThe TDT was administered after standard in-

structions were given.15 Subjects tactually exploredeach set of comparison surfaces with their pre-ferred finger and indicated the odd texture in athree-alternative, forced-choice design. Five differ-ent triplets, spanning Weber ratios of 0.033–1.0,were each presented ten times in random order toobtain the discrimination limen. The limen wasderived from fitting a cumulative normal distribu-tion to the probabilities of correct responses. In theFMT, subjects attempted to tactually match each ofthe ten test surfaces with identical comparisonsurfaces.12 Fabrics were presented behind a cur-tain, using a predetermined random order and astandard set of instructions. The preferred fingerused in the TDT was also used for the fabric test.Response and target rank orders were correlatedwith Spearman’s rho to quantify fabric discrimina-tion ability. Rho values were transformed withFisher’s z, and the scale was inverted by subtract-ing scores from the allocated maximum value of3.5 z score to make graphical presentation of ther-



apeutic change consistent in direction with TDTscores.

In the WPST for flexion-extension and ulnar-radial deviation, the examiner moved the subject’shand, via a lever, to 20 different predeterminedwrist positions. Average absolute error betweenactual and response positions was then calculatedas the index of limb position sense for each test.

SSTThe SST program was designed to maximize

improvement of the specific sensory discrimina-tions trained.8,16 Principles of training included:repeated presentation of targeted discriminationtasks (as learning is reported to be maximal for thespecific task trained); progression from easy tomore difficult discriminations; attentive explora-tion of stimuli with vision occluded; use of antici-pation trials; feedback on salient sensory featuresof the stimuli, accuracy of judgments, and methodof exploration; comparison of the sensation withthe other hand; use of vision to facilitate inter-modal calibration of sensory information; sum-mary feedback; and intensive training.8,16 Theseprinciples were applied to training each stimulus,as described below.

In the SST program, training tasks were thesame as the assessment tasks, but typically used arestricted range and number of stimuli. Five sets ofgrids, with differences in surfaces ranging from3.3% spatial increase to 100% spatial increase,were used for training texture grids, as previouslydescribed.8 Subjects explored each set of compari-son grid surfaces with their preferred finger andindicated the odd texture. Training of fabric dis-crimination used the fabrics of the FMT. Subjectswere required to match comparison surfaces to arestricted range of target surfaces and, in sometrials, to identify which fabrics they had touched.Training employing a subset of fabrics with fourlevels of difference, from large to small differences,provided a framework to assist interpretation of therelative smoothness and roughness of the set ofsurfaces. To train for flexion-extension positionsense,8 the subject’s wrist was moved to a positionin the flexion-extension range. Subjects then indi-cated perceived position of the wrist from a set ofdefined angles. Seven positions were used toachieve four levels of stimulus difference (i.e., 90-,60-, 30-, and 15-degree differences). In training forulnar-radial position sense, six positions were usedto achieve four levels of stimulus difference (i.e.,40-, 25-, 15-, and 7-degree differences).

Subjects were initially presented with the larg-est stimulus differences for which �100% accuracyhad been shown in testing during the baselinephase. Stimulus sets were explored with visionoccluded, and subjects gave a response in an at-

tempt to encourage attention to stimulus differ-ences. Quantitative feedback on accuracy of judg-ments was provided by allowing subjects to visuallycheck the defined differences (cross-modality cali-bration) and by comparing the sensory experienceusing the affected and other hand (within-modalitycalibration). Salient sensory features of stimuli(e.g., roughness characteristics of contour, spatialdensity) were highlighted by the trainer, and irrel-evant variation de-emphasized, in an attempt tosensitize subjects to the distinctive features of dif-ference. Qualitative feedback on how subjects ex-plored the stimuli (e.g., use of lateral movement ofthe fingertip over the surfaces or imagining a ref-erence point of where the hand was pointing in theproprioception task) was also given. After initialexposure to stimuli of a particular stimulus differ-ence level, a series of anticipation trials were con-ducted in which subjects were instructed that thesame stimuli set previously explored would be pre-sented. Thus, choices were limited and subjectscould make judgments with knowledge of what toexpect to feel. Subjects responded whether thestimuli were the same or different and identifiedwhich of the limited stimuli was felt. Feedback onaccuracy, sensation, and method of exploration wasprovided, as above. When correct discriminationsoccurred in at least three of four consecutive oc-casions, the next-finer stimulus difference was in-troduced. Summary feedback on the individual’saccuracy of judgments and method of explorationwas also provided at the end of each training ses-sion. Training was intensive, with sessions lasting40–60 mins, three times a week.

Data AnalysisFirst, each single-case experiment was ana-

lyzed separately for stimulus-specific and general-ized training effects. Second, these results werecombined in a meta-analysis to obtain an overallconclusion.

Time-series data were analyzed using visual(graphical) and statistical analyses.8,13 For graphi-cal analyses, standard single-case charts were eval-uated visually by the authors and a panel of threetrained, independent analysts who were naïve tothe purpose of the study. The analysts’ type I error(false alarms) and type II error (missed treatmenteffect) rates were calibrated using computer-gen-erated charts in which null or non-null effects werepresented unidentified and in random order.8,16

Analysts were required to make separate judgmentsof systematic change between phases according todefined criteria.

Statistical interrupted time-series analyses(ITSA)13 were also conducted for each single-caseexperiment. The initial step comprised model iden-tification, in which three curvilinear models of



learning likely to adequately describe the interven-tion effect were compared for goodness of fit usingthe curvilinear regression routines of SPSS. Themodels Y � a � blnX, Y � aXb and Y � aebX werecompared, consistent with earlier investigations.8

Residuals from the models were examined by au-tocorrelation and partial autocorrelation methodsto detect remaining serial dependence in the data.The best-fitting model with residuals free of posi-tive serial dependence was selected to describe thedata and perform ITSA. After investigation anddeletion of outliers, ITSA were tested for trend andlevel effects between phases to evaluate interven-tion effects, as previously described.8 A systematicdifference between phases was judged to be presentif either a statistically significant trend or leveleffect was present. Determination of a spontaneoustransfer effect involved (a) identification of an in-tervention effect on the primary trained responsetogether with (b) identification of a simultaneous,systematic change in a therapeutic direction in thetransfer response.

Finally, meta-analyses were conducted to com-plement individual subject ITSAs and assist a gen-eralized conclusion across the case experiments.The type I error probabilities of each ITSA wereconverted to z scores and then averaged to obtain apooled estimate.17 Separate meta-analyses exam-ined stimulus-specific intervention effects, sponta-neous generalization effects, and the effect of theintervention designed to facilitate generalized im-provement in texture discrimination.

Study 2: Transfer of Training AcrossTextured Surfaces Using a ProgramDesigned to Facilitate Generalization toNovel StimuliSubjects

Five further stroke patients, who met selectioncriteria previously described and had tactile dis-crimination impairment, were investigated. Pa-tients meeting these criteria were selected sequen-tially, as they presented, and gave voluntaryinformed consent.

MaterialsThe FMT was again employed, as previously

described. The Grid Matching Test (GMT)12 testedtexture-grid matching ability. The test used thesame finely graded plastic grids as for the TDT butemployed a stimulus-matching procedure, as inthe FMT. Sixteen surfaces, with spatial intervalsranging from 1500 to 3000 �m, were placed on testand comparison texture wheels and presented fortactual exploration through openings in the testapparatus. Reliability, discriminative validity, andnormative data have been obtained for this test.18

ProcedureThis study comprised five multiple-baseline,

single-case experiments. Performance on the GMTand FMT was monitored throughout the time se-ries. The first phase comprised baseline only. In thesecond phase, SST (described in study 1) was in-troduced for texture grids while the transfer re-sponse was monitored. In the third phase, SGT wasintroduced while monitoring of the untrainedtransfer response continued. Follow-up was con-ducted 12–14 wks after the end of training.

Test Procedure and ScoringThe stimulus-matching procedure was used

for both tests. For the GMT, the subject was re-quired to tactually match each of eight test sur-faces (selected from the set of 16) against theidentical set, with vision occluded. For the FMT,subjects were required to match each of the tentest surfaces with the ten comparison surfaces.Testing took approximately 10–15 mins for eachtest. Stimulus-matching ability was quantified bycorrelating the response and test values with Pear-son’s r (GMT) and Spearman’s rho (FMT). Corre-lation coefficients were then normalized with Fish-er’s z transform. A higher score reflects bettermatching.

SGTThe SGT program was designed to facilitate

transfer of training effects to untrained, novelstimuli. To achieve this, the additional principles ofvariation in stimulus and practice conditions, in-termittent feedback, and tuition of training princi-ples1,2,19 were included, as these have been associ-ated with enhanced transfer and retention. Avariety of training surfaces (e.g., paper, glass,leather, and rubber, but not fabrics) were employedin the SGT program. The surfaces comprised arange of distinctive features of roughness, includ-ing contour, surface pattern, and grit. Each differ-ent type of surface was graded from smooth torough across five stimuli and included small tolarge differences. First, large differences were in-troduced across a subset of surface types, followedby medium and fine differences. Thus, progressivegrading of difficulty was across stimuli and withinstimuli, and subjects had the opportunity of mak-ing similar discriminations across novel surfacetypes. Feedback on salient sensory features, accu-racy, and exploration method (as for SST) wasgiven intermittently rather than at every trial, andsubjects were encouraged to check the accuracy oftheir own performance. Feedback was also given onthe transfer task of identifying new distinctive fea-tures of roughness in novel stimuli. Specific tu-ition on principles underlying training, such as use



of anticipation trials and feedback, and how theseapply across tasks that the client may encounter inother environments were also included. Repeatedpresentation of stimuli, attentive exploration withvision occluded, anticipation trials, summary feed-back, and intensive training were also incorpo-rated, as for SST.

Data AnalysisCase charts were analyzed visually and statis-

tically for trend and level effects. Individual timeseries were graphed for visual analysis, and statis-tical analysis was conducted, after model identifi-cation, as previously described.

RESULTSStudy 1

Background data of subjects is detailed in Ta-ble 1. All subjects showed marked impairment onthe TDT (i.e., unable to consistently discriminatethe largest texture difference of 100% spatial in-crease [criterion of normality, �37.7% spatialincrease15]). Impaired performance was alsoevident across multiple modalities, using the fol-lowing tests and normative standards: wrist pro-prioception using the WPST for flexion-extension(criterion of normality �9.5 degrees14,16); pressurediscrimination using Semmes-Weinstein monofila-ments (normal threshold � 0.02–0.13 filament20);

hot/cold discrimination using the Roylan hot andcold discrimination kit (A629-1) (age-matched,healthy controls detect at least 9 out of 10 correct16).

Model IdentificationThe logarithmic model Y � a � blnX ac-

counted for most variance in the intervention datafor eight of the ten time series, without high pos-itive autocorrelation in the residuals of any series.Statistically significant negative autocorrelation(lag 2) was identified only in two baseline phases.The analysis of residuals thus confirmed selectionof a logarithmic model for ITSA.

Intervention Effects in the TrainedResponse

Intervention effects were clearly apparent inthree of the five first-trained primary responsesafter SST (Fig. 1, S01 to S03). These improvementswere detected above any changes between the firstand second phase of baseline of the second discrim-ination response. Improvements were marked andimmediate, with changes in performance notedafter the first training session. Performance scoresachieved by the end of training were within thenormal range15 and similar to those of the otherhand in all three cases. The panel of judges unan-imously detected a systematic change (P � 0.00) inall three series and did not identify a significant

TABLE 1 Background information of subjects in study 1

Subject

S01 S02 S03 S04 S05

Age, yrs 44 60 59 51 50Sex Male Male Male Female MaleHand dominancea Left (consistent) Right (consistent) Right (consistent) Right (consistent) Right (consistent)Affected side Left Left Right Right LeftTime since stroke

onset10.5 wks 6.5 wks 5.5 wks 13.5 wks 8 wks

Site of lesion (basedon CT scan)

Right frontoparietalinfarct

Right temporal lobe, coronaradiata, and lentiformnucleus infarct

Left frontoparietal andcorona radiata infarct

Left lentiform-internal capsularhemorrhage

Right basal ganglia,internal capsulehemorrhage

Motor function(affected UL)

Limited voluntary movement Isolated finger control No voluntary movementof wrist and fingers

No functionalmovement

No functional movement

Sensory function(affected UL)

TDT � 100PSI WPST � 23.9 degreesPressure � 6.65H/C � 5/10



TDT � 100PSI WPST � 32.7

degreesPressure � N/AH/C � N/A


Cognitive function Slight decrease in complexplanning skills, otherwise NAD

Impaired language; integrityof problem solving andmemory

Mildly reduced attentionprocesses; slowinformation processing

NAD; alert;concentration,memory, andlearning intact

NAD; attention, memory,and learning intact

CT scan, computed tomographic scan; UL, upper limb; TDT, Tactile Discrimination Test15; PSI, percentage of spatial increase; WPST, WristPosition Sense Test in flexion-extension plane14; Pressure, log10 force of Semmes-Weinstein monofilament (in milligrams) detected at preferredfinger17; H/C, hot/cold discrimination using Roylan hot and cold discrimination kit (A629-1), number of correct judgments out of ten stimulipresented to distal pad of preferred finger are reported; N/A, not available; NAD, no abnormality detected.

aHand dominance determined from Annett21 questionnaire of hand dominance.



T1

F1

parallel change in the transfer response, suggest-ing a specific training effect. Statistical analysisconfirmed both trend and level effects in the threeseries (Table 2). Improvements achieved by the endof the training were well maintained. In the re-maining two series, no intervention effect could beclaimed (Fig. 1, S04 and S05). Both series showedan improvement during baseline, limiting thescope for training effects.

Spontaneous Transfer of TrainingSpontaneous transfer of training to the second

response was not evident in two of the three casesin which an intervention effect was present in theprimary trained response (Fig. 1, S01 and S03).

Lack of a concomitant and systematic change inthe transfer responses was confirmed by the panelof judges. Although a statistically significantchange in trend was obtained for subject 3 (Table 3,S03) the change was not in the therapeutic direc-tion. Subject 2 demonstrated variable performanceon the transfer task (FMT) during sessions 1–10 ofbaseline and a reduced variability and possible im-provement during sessions 11–20 of baseline. Itwas difficult to distinguish whether these observa-tions represented continuation of the pattern es-tablished in the first baseline phase or a change inthe second baseline phase; therefore, judgment wasreserved. The panel of visual analysts failed to in-dicate a systematic change. However, a statistically

FIGURE 1 Case charts of spontaneous transfer effects within tactile and proprioceptive domains. In phase 1(sessions 1–10), performance was monitored under baseline conditions for both trained and transferstimuli. In phase 2 (sessions 11–20), stimulus-specific training was introduced to the primary trainedresponse (texture grids or flexion-extension wrist positions). Performance of the transfer response(fabrics or ulnar-radial wrist positions) was simultaneously monitored under extended baselineconditions to permit observation of spontaneous transfer effects. In phase 3 (sessions 21–30),stimulus-specific training was introduced to the transfer response and continued for the primarytrained response but at a lower intensity. Texture grid limen (PSI), score from the Tactile Discrim-ination Test in units of percentage of spatial increase.15 Fabric matching z-score, score from FabricMatching Test in Fisher’s z units.12,16 Fisher’s z scale was inverted so that graphical representationof the Fabric score was consistent with the Tactile Discrimination Test score (i.e., higher valuesrepresent poorer performance). Flex-ext position error (deg), flexion-extension wrist position senseaverage error, in degrees, based on the Wrist Position Sense Test.14 Uln-rad position error (deg),ulnar-radial deviation wrist position sense average error, in degrees, based on the Ulnar-Radial WristPosition Sense Test.12,16 S01–S05, subjects 1–5.



T2 T3

significant change in level, from 2.6 to 2.1, on thefabric-matching score was observed (Table 3). Theidentified change was in a therapeutic direction butof small magnitude and did not seem to follow thecharacteristic logarithmic function. In the remain-ing two cases (subjects 4 and 5), an interventioneffect in the first-trained primary response was not

identified; therefore, generalization of a trainingeffect was not testable.

Effect of SST on the Transfer StimuliAlthough the fabric and ulnar-radial position

stimuli did not demonstrate evidence of spontane-ous generalization effects, they did respond to SST

TABLE 2 Stimulus-specific training effects: Summary of trend and level effects from interruptedtime series analyses

Subject Time Series

Trend Effect Level Effect

t(df) P, 1-Tail t(df) P, 1-Tail

Study 1S01 Texture grids t(26) � 5.67 <0.0001 t(26) � 2.63 <0.008

Fabricsa t(15) � 3.10 <0.004 t(15) � 1.43 �0.05S02 Texture gridsa t(25) � 8.83 <0.0001 t(25) � 8.85 <0.0001

Fabrics t(16) � 2.67 <0.009 t(16) � 0.97 �0.05S03 Flex-ext t(26) � 2.51 <0.01 t(26) � 5.79 <0.0001

Ulnar-radial t(16) � 1.06 �0.05 t(16) � 4.66 <0.0002S04 Flex-ext t(26) � 3.22 �0.002 t(26) � 0.90 �0.05

Ulnar-radial t(16) � 1.85 <0.04 t(16) � 4.82 <0.0001S05 Flex-ext t(26) � 4.06 �0.0002 t(26) � 2.66 �0.007

Ulnar-radial t(16) � 1.52 �0.05 t(16) � 3.21 <0.003Study 2S06 Texture grids t(16) � 4.57 <0.0002 t(16) � 5.17 <0.0001S07 Texture grids t(16) � 2.92 <0.005 t(16) � 6.98 <0.0001S08 Texture grids t(16) � 0.09 �0.05 t(16) � 3.91 <0.0006S09 Texture grids t(21) � 3.34 <0.002 t(21) � 0.75 �0.05S10 Texture grids t(18) � 3.84 <0.0006 t(18) � 0.45 �0.05

Flex-ext, wrist position stimuli in the flexion-extension plane of movement; ulnar-radial, wrist position stimuli in theulnar-radial deviation plane of movement.

aOutliers deleted.Bold P values indicate a significant effect in a therapeutic direction; italic P values (S04 and S05) indicate a significant

difference between phases, but an intervention effect was not claimed due to pre-existing improvement in baseline. Comparisonsare between baseline (second phase of baseline for transfer stimuli) and stimulus-specific training phases.

TABLE 3 Spontaneous transfer effect to the transfer discrimination response

Subject Time Series



Study 1Tactile discriminationS01 Fabrics t(16) � 0.60 �0.05 t(16) � 0.51 �0.05S02 Fabrics t(16) � 0.38 �0.05 t(16) � 2.44 <0.01Proprioception discriminationS03 Ulnar-radial t(16) � 2.29 �0.02 t(16) � 0.45 �0.05S04 Ulnar-radial t(16) � 1.95 �0.04 t(16) � 0.93 �0.05S05 Ulnar-radial t(16) � 0.26 �0.05 t(16) � 0.59 �0.05Study 2Tactile discriminationS06 Fabrics t(16) � 1.32 �0.05 t(16) � 0.15 �0.05S07 Fabrics t(16) � 1.41 �0.05 t(16) � 1.01 �0.05S08 Fabrics t(16) � 1.20 �0.05 t(16) � 0.23 �0.05S09 Fabrics t(21) � 2.32 <0.02 t(21) � 0.23 �0.05S10 Fabrics t(18) � 0.37 �0.05 t(18) � 0.08 �0.05

Ulnar-radial, wrist position stimuli in the ulnar-radial deviation plane of movement.Bold P values indicate a significant effect in a therapeutic direction; italic P values indicate a significant difference between

phases but not in a therapeutic direction. Comparisons are between the first and second phase of baseline for the transferstimulus.



in all cases (Fig. 1). Improvements were clearlydefined and marked relative to the second phase ofbaseline. All subjects performed within the normalrange on the FMT and WPST for ulnar-radial devi-ation16 by the end of training. In the ulnar-radialseries, the lowest error scores were achieved. Inde-pendent visual analysis (P � 0.00) and statisticallysignificant trend and level effects (Table 2) con-firmed the presence of intervention effects relativeto the control baseline. Improvements achieved bythe end of training were well maintained at fol-low-up in all cases.

Study 2Background details of subjects are presented in

Table 4. Subjects showed severe deficit on the GMT(-0.12 to 0.29 z score compared with 0.62 z score,the criterion of normality18) and moderate to se-vere impairment on the FMT (criterion of normal-ity, �1.61 z score18). Most patients had the maxi-mum impairment score on the TDT. Performanceon the WPST varied.

Model IdentificationThe logarithmic model previously identified in

several of our studies8,16 and in study 1 againsuccessfully accounted for baseline and interven-tion data, leaving no statistically significant or sus-piciously high positive autocorrelation in the re-siduals of any time series.

Effects of SSTThe case charts (Fig. 2) suggested a systematic

change between baseline and SST phases for theGMT beyond any parallel change in the transferresponses of each of the subjects. Improvementswere marked, often achieving scores within thenormal performance range18 and comparable withthe other hand. These effects of SST were con-firmed for each subject by the ITSAs (Table 2, S06to S10).

Spontaneous Transfer to Untrained FabricStimuli After SST

Spontaneous transfer to the untrained FMTresponse was not evident in the second baselinephase when SST of texture grids was introduced(Fig. 2). This was clear for subjects 6–8. Subjects 9and 10 showed more variability. Lack of a transfereffect was confirmed by the results of the ITSA inall but one case (Table 3). A significant trend effectwas identified in subject 9. However, the changedid not follow the characteristic pattern achievedby training, and performance was variable; thus, aspontaneous transfer effect was not claimed.

Transfer to Novel Textured Stimuli AfterSGT

Discrimination of fabric surfaces, which hadnot been trained in any of the three phases, im-

TABLE 4 Background information of subjects in study 2

Subject

S06 S07 S08 S09 S10

Age, yrs 63 88 70 65 47Sex Female Female Male Male MaleHand dominancea Right (consistent) Right (consistent) Left (inconsistent) Right (inconsistent) Right (consistent)Affected side Right Left Right Left RightTime since stroke

onset5 wks 4.5 wks 13 wks 52 wks 18 wks

Site of lesion (CTscan)

Left striatocapsularinfarct

Right posterior parietaland basal gangliainfarct

Left posterior limbinternal capsularhematoma

Infarct in rightoccipital and parietal(superior) lobes

Extensive lefttemporoparietal andcorona radiata infarct

Motor function(affected UL)

No active movement Limited activemovement;

Preexisting tremor

No functional use(flicker only)

Independent fingermovement;

clumsy

Independent fingermovement

Sensory function(affected UL)

GMT � 0.15FMT � 0.56TDT � N/APSI WPST � N/A

GMT � 0.05FMT � 0.12TDT � 100PSI WPST � N/A

GMT � �0.12FMT � �0.16TDT � 100PSI WPST � 34.4

degrees

GMT � 0.18FMT � 0.99TDT � 100PSI WPST � 11.6

degrees

GMT � 0.29FMT � 0.89TDT � 100PSI WPST � 13.2

degreesCognitive

functionNAD NAD Expressive dysphasia NAD Dysphasia;

mild visuospatialproblems

CT scan, computed tomographic scan; UL, upper limb; GMT, Grid Matching Test (Fisher’s z scores)12; FMT, Fabric Matching Test (Fisher’s zscores)12; TDT, Tactile Discrimination Test15; PSI, percentage of spatial increase; WPST, Wrist Position Sense Test in flexion-extension plane14; N/A,not available; NAD, no abnormality detected.

aHand dominance determined from Annett21 questionnaire of hand dominance.



T4

F2

proved on introduction of the SGT program in fourof the five experiments (subjects 6–9), as shown inFigure 2. Improvements were often marked,

achieving scores within the normal performancerange.18 They followed a pattern similar to the SSTeffects, although the magnitude of early change

FIGURE 2 Case charts of stimulus-generalization training effects in the tactile domain. In phase 1 (typicallysessions 1–10), performance was monitored under baseline conditions for both trained and transferstimuli. In phase 2 (typically sessions 11–20), stimulus-specific training (SST) was introduced to thetexture grids and spontaneous transfer effects were monitored in the transfer-fabric stimulus. Inphase 3 (typically sessions 21–30), stimulus-generalization training (SGT) was introduced, andmonitoring continued for the fabric response to permit observation of transfer of training effects.Grid-matching and fabric-matching test scores are in Fisher’s z units.12,16 In contrast to Figure 1,higher values represent better performance. Case chart S06:trained/untrained grids provides anexample of the case charts in which the grid-matching score was separately calculated for grids thatwere trained and those that were not trained. Stimulus-specific training effects are evident in thetrained grid score, and spontaneous transfer of training effects are evident in the untrained grid score.S06–S10, subjects 6–10.



was often smaller. Subject 10 also showed an im-provement trend in phase 3; however, this was notsignificantly different from the preexisting trendevident in phase 2 (Table 5). Statistical analyses aresummarized in Table 5.

Spontaneous Transfer from Trained toUntrained Grids After SST

To investigate whether transfer might occuracross stimuli with the same texture characteristic,we trained for only half of the grid surfaces, whilemonitoring performance on all (Fig. 2, S06:trained/untrained grids). Spontaneous transferfrom trained texture grids to untrained texturegrids did occur with the SST program in four of thefive case experiments (Table 6). Subject 10 was theonly subject who did not show either spontaneoustransfer to stimuli with the same distinctive feature(texture grids) or facilitate transfer to novel stimuli(fabrics).

Meta-analysesA meta-analysis investigating the overall effect

of SST with the plastic texture grids was performedon the seven subjects providing appropriate data(subjects 1, 2, and 6–10). This meta-analysis veryclearly confirmed the positive effect of training onthe discrimination of texture grids in both level (Z� 7.61, P � 0.001) and trend (Z � 8.46, P � 0.001)variables. Discriminations within the same diffi-culty range, but using plastic grids with spatialintervals that were not used during training trials,were available from five experiments (subjects6–10). This meta-analysis also showed significantlyimproved performance in both the level (Z � 3.43,P � 0.001) and trend (Z � 4.61, P � 0.001)variables. These strong positive results in discrim-inations of trained and untrained texture gridscontrast markedly with the results obtained forspontaneous transfer–related changes in the dis-crimination of fabric textures after SST of plasticgrids. Meta-analysis of these time series failed toshow a statistically significant effect in either level(Z � -0.206, P � 0.41) or trend changes (Z �

0.126, P � 0.55), despite the increase in sensitivityafforded by the pooling of seven sets of experimen-tal results. A meta-analysis that directly contrastedthe effect of specific grid training on discrimina-tion of plastic grids vs. discrimination of fabricsshowed a clearly superior effect in the discrimina-tion of the plastic grids according to both level (Z� 5.24, P � 0.001) and trend (Z � 6.07, P � 0.001)variable results. Importantly, when the SGT pro-gram was introduced in the second group of fiveexperiments, both level (Z � 4.10, P � 0.001) andtrend (Z � 5.57, P � 0.001) variables respondedwith significant changes in the untrained transferstimuli.

DISCUSSIONSeveral conclusions seem reasonable. First,

rapid, task-specific training effects can be obtainedwith the SST program. Second, improvements ob-tained in a particular discrimination task do notgeneralize spontaneously to related stimuli withinthe same sensory-perceptual dimension in strokepatients after SST. For example, training on gridsurfaces did not generalize to fabrics. However,spontaneous transfer of training does seem to oc-cur across stimuli that share the same distinctivefeature (e.g., trained and untrained grids) afterSST. Third, performance on related untrainedstimuli within the same sensory-perceptual dimen-sion can be improved to a clinically significantdegree by using a program deliberately oriented toenhance generalization (the SGT program). Trans-fer of training from practice with paper, glass, andrubber to discrimination of fabric surfaces illus-trates this point.

Conditions of Training and Influence onTransfer

Successful transfer across stimuli seems tohave been influenced by the training principlesemployed. The SST program employed principlesdesigned to maximize learning to discriminate aparticular stimulus type. Positive training andtransfer effects within the specific stimulus type

TABLE 5 Stimulus-generalization training effect: Transfer to novel textured stimuli

Subject Time Series



S06 Fabrics t(16) � 5.37 <0.0001 t(16) � 3.76 <0.001S07 Fabrics t(16) � 3.41 <0.002 t(16) � 1.59 �0.05S08 Fabrics t(16) � 4.13 <0.0004 t(16) � 2.22 <0.02S09 Fabrics t(21) � 1.77 �0.05 t(21) � 2.10 <0.02S10 Fabrics t(23) � 1.13 �0.05 t(23) � 1.31 �0.05

Bold P values indicate a significant effect in a therapeutic direction. Comparisons are between the second phase of baselineand the stimulus-generalization training phase for the transfer stimulus.



T5

T6

trained were achieved. However, spontaneoustransfer to related tasks within the same sensory-perceptual dimension did not occur. To facilitategeneralized improvements, a program that addedvariation in training stimuli, intermittent feed-back, and tuition of training principles was re-quired. Repeated exposure under conditions of at-tentive exploration with vision occluded wasusually insufficient for therapeutic change, as in-dicated by stable baseline performance in 75% ofthe baseline time series investigated. Thus, condi-tions of training seem important to outcome, withimplications for the design of efficient rehabilita-tion programs. The principles of training derivedfrom studies of learning with normal subjects seemapplicable to stroke patients, confirming the valueof learning theory for models of neurologicrehabilitation.

Specificity of Learning and InformationProcessing Within the Sensory System

Perceptual learning studies with unimpairedsubjects suggest that discrimination training is of-ten highly specific to the task and receptor locationand to the method of processing information.2,3 Wealso found highly specific training effects in tasksemploying the same sensory dimension (tactile orproprioceptive) and body location (fingertip orwrist) when SST was undertaken. Spontaneoustransfer was, however, found across stimuli of thesame type that employed the same receptor loca-tion and method of processing the information(e.g., from trained to untrained grids and acrosswrist positions within the same plane of movement[positions trained were different to those as-sessed]). These findings suggest spontaneoustransfer of training to novel stimuli of the sametype and of comparable difficulty when SST isemployed.

The high degree of specificity observed is alsoconsistent with evidence of highly specific deficitsresulting from cerebral damage.12,21,22 Moreover,studies of behaviorally induced neural plasticitysuggest that changes in cortical maps are specificto the trained task and specific body location inmonkeys23 and humans.24 Organization of the so-matosensory system is highly specific, as evidencedby body location and modality-specific columnarorganization and submodality-specific neurons inthe primary somatosensory cortex and neuro-ax-is.25 These observations indicate a high degree ofspecificity of information processing within modal-ities and body locations. However, the sensory sys-tem is also characterized by convergence of so-matosensory information26 and presence ofdistributed sensory networks.25 Thus, distinctivesensory features of a multidimensional texturemight be integrated in a unified perception at alevel of convergence in the system.26 This level ofprocessing would be consistent with the learningand transfer across textures with multiple featuresof roughness after SGT.

What Is LearnedA common view supported by behavioral2

and neurophysiologic27 evidence proposes thatdistinctive features of difference are learned andform the basis of transfer of training. In the TDTand GMT, the distinctive feature is likely to bethe spatial interval of the grids,27 whereas thephysical characteristics of fabrics are likely to bebased on nonspatial neural coding mecha-nisms.27 This inference is supported by observa-tion of stimulus-specific transfer within, but notacross, these stimuli. The distinctive feature ofthe limb-position task is presumed to be thechange in location of the limb28 or relative wristangle. Our findings suggest that the distinctive

TABLE 6 Stimulus-specific training effect on trained grids and spontaneous transfer to untrainedtexture grids

Subject Texture Grids



S06 Trained t(16) � 5.49 <0.0001 t(16) � 6.66 <0.0001Untrained t(16) � 3.02 <0.004 t(16) � 1.34 �0.05

S07 Trained t(16) � 1.49 �0.05 t(16) � 4.70 <0.0001Untrained t(16) � 0.62 �0.05 t(16) � 3.21 <0.003

S08 Trained t(16) � 0.67 �0.05 t(16) � 2.72 <0.0002Untrained t(16) � 0.41 �0.05 t(16) � 2.31 <0.02

S09 Trained t(21) � 4.77 <0.0001 t(21) � 1.08 �0.05Untrained t(21) � 5.04 <0.0001 t(21) � 0.12 �0.05

S10 Trained t(18) � 5.30 <0.0001 t(18) � 0.06 �0.05Untrained t(18) � 0.40 �0.05 t(18) � 0.66 �0.05

Bold P values indicate a significant effect in a therapeutic direction. Comparisons are between baseline and stimulus-specifictraining phases for the trained grids and between the first and second phase of baseline for the untrained grids.



feature also needs to be defined in relation to thedefined plane of movement or to receptors in-volved in the discrimination. This conclusion isconsistent with observations indicating selectiveactivation of muscle receptors,29 joint receptors,and skin areas in association with specific posi-tions.28 Thus, it seems necessary that the distinc-tive features of stimulus difference are highlysimilar and specific to the receptor populationactivated to obtain spontaneous generaliza-tion.

Using the SGT program, transfer was obtainedacross tactile stimuli with potentially different dis-tinctive features (i.e., from training on rubber,glass, leather, and sandpaper to transfer on fab-rics). Training of multidimensional stimuli with awide range of distinctive features in the SGT pro-gram is likely to increase the probability of expo-sure to distinctive features relevant to the transferstimulus, with consequent transfer of learning. In-vestigations with unimpaired subjects have sug-gested that transfer is facilitated when stimuli aremore complex and potentially share some distinc-tive features.2 Physical characteristics of differencetrained in our study included surface type, surfacecontour, grit, friction/abrasion, stimulus patternswith different spatial frequencies, surface irregu-larities, and various combinations of these. This setof features seems to represent a broader dimen-sion2 of roughness and covers aspects of roughnessthat are both spatial and nonspatial. Thus, successof the SGT may be influenced by the fact that thevaried stimuli used included nonspatial discrimi-nations that may be more directly related to thetransfer stimulus. Further, it may be the dimen-sion of roughness, rather than a specific unitaryfeature of it, that is learned and transferred in thistraining. Other studies obtaining generalized train-ing effects in stroke patients9–11 have includedvariation in training stimuli, supporting thisinterpretation.

Generalization of Training Within aDisordered Somatosensory System

The ability of stroke patients to generalizelearning to novel tasks is relatively unknown, par-ticularly within the somatosensory domain. Strokepatients with somatosensory impairment may nothave the capacity to process the intrinsic somato-sensory input from the untrained transfer task ormay have a more global difficulty in generalizinglearning effects. Our findings confirm both task-specific and generalized transfer effects. The ob-served effects suggest that stroke patients cantransfer learning effects, but the boundaries oftransfer do have limits. More importantly, theysuggest a high degree of specificity of learning and

provide some guidelines on what features of thetask need to be similar before learning transfers.

Subjects with an unimpaired sensory systemcan improve perceptual discriminations with prac-tice alone and demonstrate spontaneous perceptualtransfer.3 In contrast, stroke patients in the presentstudies repeatedly practiced with the assessmentstimuli during baseline, yet in the majority ofcases, they did not demonstrate improvement onthese stimuli. This finding highlights the needfor appropriate conditions in rehabilitation training.Repeated exposure alone, which approximatesexposure to sensory stimuli in daily activities,was not adequate to effect a clinically significantimprovement.

After damage to somatosensory cortical areasand their connections, it is likely that stroke pa-tients have an inadequate foundation from whichto perceive and discriminate novel stimuli, al-though they may have learned the principles ofhow to process similar types of information better.Perceptual learning at any given time is con-strained by the existing structure.2 Evidence thatlearning involves modification of representationsin the brain23,24 suggests that perceptual trainingis not only a means of training the process ofperceptual learning. Individuals must have accessto the distinctive characteristic of the stimulus toextract the distinctive feature. External quantita-tive feedback on what is being perceived may benecessary to make sense of or to “calibrate” per-ceptions in a changed (e.g., after brain damage)system.28

Study LimitationsInvestigation of training effects was limited to

ten subjects (20 time series). Although this is asmall number for group-based studies, each inten-sive single case is a controlled experiment involv-ing within-subject control and replication of train-ing effects. This approach is well suited to theheterogeneous nature of stroke and allows for vari-able findings across individuals that might be ob-scured with group analysis. In addition, systematicreplication of training effects across ten subjectswith different background characteristics andacross tactile and proprioceptive modalities pro-vides generalizability of findings not usually af-forded to single case studies. Finally, meta-analysisof individual experiments improves power, permitsan overall conclusion, and provides a quantitativeevaluation of the generality of treatment effects.13

Thus, our findings provide a strong foundation forlarger controlled studies.

The sequence of baseline, SST, and SGT wasconstant, and therefore, it may be argued that theexposure and SST phases led to an overlearningeffect with an impact on transfer and effectiveness



of the SGT program. However, this explanation isunlikely given that observed SST effects workedquickly, that even prolonged exposure (e.g., 20sessions) did not effect changes, and that changesobserved for SGT were of similar rapidity. Theseobservations support the conclusion that improve-ments in the SGT phase were due to a generaliza-tion training effect, rather than due to additionaldosage of stimulus exposure, or SST.

Implications for TherapeuticInterventions and Further Investigations

Clinically significant improvements were ob-tained with training, providing justification fortherapeutic intervention. Practice or exposurealone was not usually sufficient to achieve thechanges characteristic of perceptual learning. Thefindings also suggest that the nature of training iscrucial to outcome. SST may be important if thereis a need to train specific work or daily living tasks.However, to facilitate more generalized improve-ments and minimize reliance on resource-inten-sive, task-specific training, a program that alsoincludes variation within and across related train-ing stimuli, intermittent feedback, and tuition oftraining principles is recommended. In the major-ity of cases, ten training sessions seem to be suffi-cient to achieve clinically significant improvement.However, the goal of healthcare providers is notonly to provide an effective training program butalso an efficient training program. The highly spe-cific nature of training suggests that cliniciansneed to accurately identify the specific sensory di-mension that requires training and use gradedstimuli that cover the range of distinctive featuresfor the perceptual dimension targeted. Programscannot rely on spontaneous transfer; they need tosystematically train for transfer. This issue of gen-eralization of training effects has importance inrelation to rehabilitation of a range of abilities instroke patients and to rehabilitation in general.

The present studies provide some guidance onboth the nature of transfer expected and on theconditions of training required to achieve transfer.Further systematic investigation of these variablesis indicated. For example, transfer across macro-spatial and microspatial features of surfaces willhelp to determine if the boundary of spontaneoustransfer is at the level of a particular spatial patternor broader spatial features of roughness. Transferacross dimensions such as smoothness–roughness,hardness–softness, and compressional elasticitywithin multidimensional texture stimuli is also in-dicated, particularly after SGT. Investigation oftransfer across different joints of the upper limb forproprioception or different fingers for texture isalso necessary. Identification of sites of cerebralreorganization underlying the recovery of somato-

sensory function after stroke30 may provide newinsights into cerebral networks involved in percep-tual learning and recovery. Systematic investiga-tion of the relative contribution of training princi-ples in the treatment package would also help toidentify the critically important elements associ-ated with successful learning and transfer. Finally,the ability to modify sensory abilities experimen-tally opens up an experimental paradigm for futureinvestigations of the relationship between sensa-tion and other abilities, such as pinch grip, and theeffect of improving sensation on motor function.

ACKNOWLEDGMENTSWe thank the participants for their involve-

ment in the study and the therapists and manage-ment at Royal Talbot Rehabilitation Centre, CedarCourt Healthsouth Rehabilitation Hospital, andHampton Rehabilitation Hospital, Victoria, Aus-tralia, for their support.

REFERENCES1. Schmidt RA, Lee TD: Motor Control and Learning: A Behav-

ioral Emphasis. Champaign, IL, Human Kinetics, 1999

2. Goldstone RL: Perceptual learning. Ann Rev Psychol 1998;49:585–612

3. Sathian K, Zangaladze A: Tactile learning is task specific buttransfers between fingers. Percept Psychophys 1997;59:119–28

4. Nagarajan SS, Blake DT, Wright BA, et al: Practice-relatedimprovements in somatosensory interval discrimination aretemporally specific but generalize across location, hemi-sphere and modality. J Neurosci 1998;18:1559–70

5. Ahissar M, Hochstein S: Task difficulty and the specificity ofperceptual learning. Nature 1997;387:401–6

6. Spengler F, Roberts TP, Poeppel D, et al: Learning transferand neuronal plasticity in humans trained in tactile discrim-ination. Neurosci Lett 1997;232:151–4

7. Zihl J, Von Cramon D: Restitution of visual function inpatients with cerebral blindness. J Neurol Neurosurg Psychi-atry 1979;42:312–22

8. Carey LM, Matyas TA, Oke LE: Sensory loss in strokepatients: Effective tactile and proprioceptive discriminationtraining. Arch Phys Med Rehabil 1993;74:602–11

9. Yekutiel M, Guttman E: A controlled trial of the retraining ofthe sensory function of the hand in stroke patients. J NeurolNeurosurg Psychiatry 1993;56:241–4

10. Byl N, Roderick J, Mohamed O, et al: Effectiveness of sen-sory and motor rehabilitation of the upper limb followingthe principles of neuroplasticity: Patients stable poststroke.Neurorehabil Neural Repair 2003;17:176–91

11. Smania N, Montagnana B, Faccioli S, et al: Rehabilitation ofsomatic sensation and related deficit of motor control inpatients with pure sensory stroke. Arch Phys Med Rehabil2003;84:1692–1702

12. Carey LM: Somatosensory loss after stroke. Crit Rev PhysRehabil Med 1995;7:51–91

13. Franklin RD, Allison DB, Gorman BS (eds): Design andAnalysis of Single-Case Research. Mahwah, NJ, LawrenceErlbaum Associates, 1997

14. Carey LM, Oke LE, Matyas TA: Impaired limb position senseafter stroke: A quantitative test for clinical use. Arch PhysMed Rehabil 1996;77:1271–8

15. Carey LM, Oke LE, Matyas TA: Impaired touch discrimina-



tion after stroke: A quantitative test. J Neurol Rehabil1997;11:219–32

16. Carey LM: Tactile and Proprioceptive Discrimination LossAfter Stroke: Training Effects and Quantitative Measure-ment [PhD thesis]. Melbourne, LaTrobe University, 1993

17. Rosenthal R: Meta-analytic Procedures for Social Research,Revised Edition. Newbury Park, Sage Publications, 1991

18. Carey LM, Matyas TA, Oke LE: Quantitative measurement oftactile and proprioceptive discrimination in the upper limbafter stroke. Stroke 2000;31:2816–7

19. Cormier SM, Hagman JD: Transfer of Learning: Contem-porary Research and Applications. San Diego, AcademicPress, 1987

20. Directions for Using Semmes-Weinstein Monofilaments.San Jose, North Coast Medical, 1987

21. Annett M: A classification of hand preference by associationanalysis. Br J Psychol 1970;61:303–21

22. Roland PE: Somatosensory detection in patients with cir-cumscribed lesions of the brain. Exp Brain Res 1987;66:303–17

23. Recanzone GH, Merzenich MM, Jenkins WM: Frequencydiscrimination training engaging a restricted skin surfaceresults in an emergence of a cutaneous response zone incortical area 3a. J Neurophysiol 1992;67:1057–70

24. Karni A, Meyer G, Jezzard P, et al: Functional MRI evidencefor adult motor cortex plasticity during motor skill learn-ing. Nature 1995;377:155–8

25. Mountcastle VB: The columnar organization of the neocor-tex. Brain 1997;120:701–22

26. Mesulam MM: From sensation to cognition. Brain 1998;121:1013–52

27. Johnson KO, Hsiao SS: Neural mechanisms of tactual formand texture perception. Ann Rev Neurosci 1992;15:227–50

28. Clark FJ, Horch KW: Kinesthesia, in Boff KR, Kaufman L,Thomas JP (eds): Handbook of Perception and HumanPerformance, chapter 13. New York, John Wiley, 1986

29. Gandevia SC, McCloskey DI, Burke D: Kinaesthetic signalsand muscle contraction. Trends Neurosci 1992;15:62–5

30. Carey LM, Abbott DF, Puce A, et al: Reemergence of acti-vation with poststroke somatosensory recovery: A serialfMRI case study. Neurology 2002;59:749–52

balt5/z79-phm/z79-phm/z7900505/z792796-05a browncm S�10 3/11/05 4:48 Art: 162531 Input-mbb(v)


training of somatosensory discrimination after strokeof somatosensory deﬁcits, particularly touch...

Documents