Kauppinen P, Kööbi T, Kaukinen S, HyttinenJ, MalmivuoJ. Application ofcornputer modelling and lead field theory in developing multiple aimed impedancecardiography measurements. Journal of Medical Engineering dz Tuhnohgt. in press

Reprinted v/ith permission ftom TÅYLOR & FRANCISLtd5http:/ /www.tandf.co.uk

Papen MET 100343

Jomal of Mediel EoginsiDg & Technologr, Volue fi1, Nuber 00, (llln?? l9tt9), pag6 000-000 ,wApptication of computer modelling and leadfield theory in developing multiple aimedimpedance cardiography measurernents

P- I{auppinent, T. Kööbil, S. Kaukinen$, Since ICG measurements are not focused to arry

J. Hyttinenf andJ. Malmivuof specific activity of tle thorax, the contributions to Å7

lTutpae Univmitj oJ Tehnolog, Ragw &ant In*iaaa P.O. Box 592, are not known exactly, rendering it difficult to extractFN-3t101 TamPaeFmlot{ruiLhll372l@ehtf - specific information to qgantifu cardiovascular para-lTmpae Uniuaig Hq?ital, Dcpotmt oJ CIini.aJ Wg^, Tqnqm, Å.t .., such as cardiac output (CO). Conditioru iuchfrnbnn /h>'6'o t's'J

STcilen hiunt) Hvpital Deputwt of Anattwia and Intmivc &ta 3 ,T5ttT*":. tlPo and ralvular disease might be

r" i*,ria-ra' indistinguishable [11, 12]. Also, various anatoniical and

conaentionatimped.ance cardiographl ecG) nzthnds 6tinat' *ä::'ffiffi:'ffft 5ffiilä*riliJ3,*.-"tporaratzrs relatzd tn thz function of tlw hean frnn a singleuoueform th.at rqlzcts an inkgrated, cottbirtntiml of compbx ICG has mainly been snrdied by clinical evaluation,sources. We ha,ae praiously dzaelapd methods and, took for while few studiei have utili"ed moäeb of the thorax as acahula,ting .nzasnarLent sensitittily tlisnibutions of ICG volume conductor t7 , I , 12, 1 3l . Yet, these studies haveebctrcile configtnatiors. In this stutly, tlw mzthads uae generally not provided insight into improving the basisappliztl to inuestigatz tlu pmspeds of recording mula'pb aincd öf the method. The applica6ility of the lead fi-eld theoryICG waaefonu urilizing th4 12-.lad ebdrocard'ingraplry in impedance me"sroC-.trt" has been shown theoreti-(ECG) ebmodz locatinns. Tlwee atntamical$ realistit uolume caly by Geselowitz in 19?1 [14]. Based on that theory,conhrxtm rnod4ls were tueL onz based on. Wsibb Htman Man by åppiopriately selecting the electrode configurarion itayosection data and, tuo on nngnztix rfon-anre (\t!) lma-fs is p6siiUtä to anite ar meas'rement imposingincreasedrepresarting end diasulir anil m.d systnlic pha^sa of thz cgrd',gc -.o -ä"sur.ment sensitivity and selectivity to particularqch. Bosed on tlu sensiti.vit, d,istrihtioru obtainzd, BA'rt regions. Selective measurements could improve theelzctroilz confignatianu were selzcted' for prain_lnyy aen(l re[ability of ICG by offering infonnation ielated toacamhntim on 12 leahhy aoluntzsrs anil 9 ztalaular 6sgtizc ftrnction oi abnoråalities not detected byPatiilLts. The moilzl sadl sugateil that a ua.rizty of conventional methods. Since multiple sources contri-configwatitns had clur\ enhanced. sensitivi$ to tlu card;io- bute to

^7, an approach where the quantity of signals is

aasailar stnrctures as compared, n corumtimtal ICG. recorded nighf 6e useftrl in estimådng cardia- para-Sirnil"atioL datn and, clinital acpaiments shuted' Iogiral meters.cott84ondance suppofiing tlu thuretital$ pedictzd' d'ifferences

betlr)en tfu conf,grrations. Recutdzd L2-lzad ICG signab had There exist no srudies in which theoretical knowledgechararteristic and lnndmat'hs not coincid;ing atith of the forrnation of ICG measurement sensitivity hasthose of conaentional ICG. hnthcrtnorc, conf.gtatiotts been applied in developing ICG electrode configura-shouting raenblance to irwasiae il'ata and noQhnlogical tions. Itecent work has inad'e ir possible to analpö theaarintit Ls in dbease are of i.nwest. Tfu results in"dbate thz measurement sensitivity distributions of ICG using theWlitabiliry of thc nadzlling alptoach in dateloping- IlG lead field theoretical' approach with computerizedmzavfiatwfi configtnations. Houeuo, tlv leoel of cliniral vol'me conductormodelling [9, 10, 15, 16]. Simulatedrebuante anil potential of th.e I2-lcad mzthad renains ta be sensitivity distributions can be used to approximate how*plmed in stud,izs mploying dynamir nodzlling and' conductivity changes in different regionJ of the humanacquisitian of inaasiue refnewe d,ata. *rorax affect the measured impedance signal.

Inhoduction The objective of this study was to investigate thepossibility of developing multiple aimed ICG measure-

Conventional ICG techniques provide a single impe- ment configurations with enhanced detection of thedance u-acing, from which paramet€rs related to the heart-related components in the recorded Å2. Volumepump function of the heart are estimated I I - 4] . Time conductor models of the human thorax were applied tovarying changes in impedance (ÅZ) reflect the inte- simulate the sensitivity disributions of such impedancegnted cornbination of multiple sources, including measurements as can be established using the l2-leadtissue volume and movement, tissue resistivity, blood ECG elecrode system. The 12-lead electrode qrstem wasdisaibution and blood flow changes [5-7]. The selectedsinceitprovidesadinicallyeasilyadoptedbasiscontributions of these phenomena are reflected in Åz for new measurement method with the possibility ofdepending on the electrode configuration used for the simultaneous ECG recording. Based on computermeasurement [5, 6, 8-10j. simulations, electrode configurations producing either

ld oJ Mtdtu'I EBgi'EäE fr' IäoQg ISSN 03091S02 ptut/ISSN r4d!522( onlisc @ 1999 Taylor & Fdcis IldhEp://w9w.hdf.@.u&lfNls/ncrhh

hapr,//w.aylomdtucis.åE,4NIJ/mcLhb

P. Xauppino ei al Dweloping multiple ICG meruements

high sensitivity or selectivity in larious tissues andorgans were chosen for preliminary clinica.l examina-tion on volunteers and valvular patients. Generalmorphology and inter-individual variabfity in recordedsignals were considered and compared with thetheoretical results obtained.

Materials and methods

ICG smsitivity distrihdiorz

ICG sensitivity distribution describes the ability ofapplied electode configuration to detect conductivitychanges within the body. The measured basal impe-dance 7a and its change ÅZ resulting from theconductivity o and its change Äo within a volumeconductor can be evaluated by

( )z:[ -1-s,.. (

J" ra)"t' (1)

where S(x,y,z,l) is the scalar fietd giving the sensitivity toconductivity changes at each location. S can beobtained by determining two independent currentfields generated by a unit current applied to thecurrent injection electrodes and the voltage measure.ment electrodes. These fields form the lead fieldsassociated with tåe electrode setting. The sensitivityfield S is the dot product of the a,t/o fields:

s :Ju.Ju (2)

where ../rr is the lead field produced by currentexcitation electrodes, nd Jrn the lead field producedby the reciprocal energization of voltage measurementleads [4,17].

{Jrilizing equation I witl finite difference method(FDM) computer modelling, information as ro therespective capacity of diffierent ICG measurements todetect conductivity and ia changes in the thoran< ren beestimated t9, l0l. In FDM, the modelled volume isdivided into a three-dimensional resistor nenvork whichreflects the thorax both geometrically and as aconductor. Methods to consEuct and solve accuratevolume conductor comDuter models based on FDMhave been previously developed and validated [15, 18 -201.

The relative magninrde of the sensitivity field in a tissueqpe (or a group of tissues considered as one targetvolume) gives a measure of how conductivity lariationin that tissue will afect the detected M. The overallsensitivity of a tissue type is obtained by integrating thesensitivity lalues of the tissue over the volume itoccupies. This sensitivity value can then be comparedwith the absolute total sensitivity of the model as givenbv

.z nt

+ 7007o (3)

where z" is the number of FDM elements in the targetvolume znd n, the number of tissue elements of acertain rype. The denominator is the sum of the

absolute partial contributions from all tissues (or tissuegroups), and the numerator is the contribution of thetarget tissue.

Volumc conductor modzls

Three dif[erent models were emploved. Two modelsrepresenting end diastole (end 'diastole

model

-EDM) and end qntole (end qntole model-ESM)were constructed from two ECG triggered magneticresonance image sets obtained &om a healthy malesubjecr Image data has been described earlier [7].The third model was based on high resolutioncryosection anatomy data &om the US NationalLibrary of Medicine's Visible Human Man (visiblehuman model-WlM) project t211. All råree imagesets were segmented to 26 tissue types and organsbased on a segmentation algorithm that provideselements for the FDM caicr.rlation [22]. The resolu-tion of FDM elements in the ECC-triggered modelsvaried from 0.10 to 5.8 cm", resuhing in 121 431elements. For the \rHM, the resolution was from0.044 to 5.7 cm', compriiing 162 786 elemena [16].Table I presents selected tissue voh'-es of themodels and volume changes between EDM andESM as well as between the VIIM and the averasevolumes of EDM and ESM.

Doiuation and ana$sis of L2-lzad, ICG configu.rations

The nine electrode locations of the l2-lead ECGelectrode system were used separately to calculate abasic set of lead fields for each model. A computeralgorithm was developed to make possible combina-tions with the l2Jead electrode sptem using atmaximum four electrodes at a time for either leadfreld in equation 2. Configurations, which utilize thesame electrode location for current injection andvoltage measurement, were omitted to reduce theskin-electrode impedance effect on Å2. Deriving ICGmeasurement combinations with the ore-calculatedlead fields is a simple non-iterative calåIation, sinceöe sptem is assumed to be linear. For example, a leadfield betrseen the chest leads Vl and V6 may beobtained by subtracting V111 from V6g. On the otherhand, the same result is obtained by subtracting \llsfrom VGs. A total of 65 476 impedance measurementconfigurations utilizing the l2-lead electrode locationswas thus derived.

A database was computed for each model and 65 476measurement configurations containing the informa-tion on the formation of Ze and proportional contribu-tions according to equation 3. This was done for eachtissue listed in Table I and for a number of differenttissue groups reflecting functional structures of thecardiorzscular system. Trssue groups were formed e-g.from the tissues forming the sptemic and pulrnonarycirculation in addition to groups containing smallernumber of tissues such as left aaä together with leftventricle. Further, the same calculations were applied tothe data produced by subtracting the sensitivity and Zeyalues simulated by the ECGtriggered modeis EDMand ESM.

P. Kauppiao et ol Doeloping multiple ICG meauemcnts

Tabb 1- Volurc of selzctzÅ tiswa in appäcd mdzb i,n ad"dition tn uoluru changes behnem tlu ad diastolz and. nd ststtb modzls and benrm the

uisibb hum modzl anl naage of trigmd, modzb.

vHM tll EDM tll EsM UlLY Vo ÅV % (VHM to avg

(EDM-ESM) of EDM+ESM)

Hean muscleHeart fatLeft atriumRight ariumLeft venuicleRight ventricieAscending aortaAonic archDescending aortaSuperior vena calaInferior vena carraPulrnonary arteryPulrnonary veinCarotid ateryJugulr veinOther bloodleft lungRight hmgSkeletal mrudeFatTotal volme

0.310.200.0120.110.0r00.0s70.0I30.0r80.0310.0220.0910.0380.0270.0070.0390.0771.71.8r8.816.7i16.9

0.290.120.0340,0540.0950.0150.0320.0210.0590.0180.0290,0840-0310.0r20.0r80.0271.01-1

6.29-C

2t.2

0.290.140.0520.0820.0530.0620.0360.0270.0690.0180.0300.r50.0150.0140.0250.0431.0tl6.35.4

27.2

1.9

-12-35-34

80t43-tl_19

-14-0.6-3.4-44100

-18-30

3.11.8

- 1.0l.l

95l

- lJ55

-86_J

-6r

-6991

209

-681l/Å

80120696

200206

The most sensitive configurations for classified tissuesand defined tissue groups were selected from thedatabase. Configurations for rhe same single tissue andtissue groups with minimal simultaneous sensitivity inthe lung regions were also separated. Additionallnconfigurations producing high sensitivity in the lungswere selected, resulting altogether in 237 öfferentelectrode configurations.

Ex\dittcntol cknicalmzasneinenu,lnAn experimental pilot sordy was coliducted to evaluatethe impedance waveforrns detecpd by the configura-tions selected based on the sitnulations. The studyinvoived welve healtly volglteers (age 30.5+6.4ymean*SD, range 20-42 y,fr male 3

-female, weigtrt79+ l6 kg, 55-100 kg; height 179+8.0 cm, 164-192 cm, BMI 24+3.6 kg/lrl.', 17-30 kg,/m=). Themeasurements were also taken preoperatively on agroup of nine patiena with valwlar heart disease (threemitral, six aortic: age 58.4+9.6, 35-72y; all male;weight 76+14kg, 62-i1l kg; height 171 +2.8 cm,166-175 cm, BMI 26+5.2 kglm',20-39 kglm').

The measurements were Derformed bv CircMonw8202 0R medical Ltd, 'Tallinn,

Estonia), whichincludes an impedance channel delivering 0.7 mA at30 kHz. A novel software{ontrolled swirching devicecapable of electrically connecting the irnpedancemeasurement terminals to any number of appliedelectrodes was used in combinatioan with theCircMon [23]. The electrode configuration used forimpedance measurement could thus be alteredrapidly by computer control without manual opera-tion.

)( Al{ ao-a.-a4nq,L.'were +4Yz^ q\41z-+biectj lr <upiqe fa<.i-D^ Äreaih"r.1å ia; -. ̂ r ?öL!-a i-äs . a

All subjects were prepared for the procedure byattaching disposable ECG spot electrodes (type QIGAmanufacarred by MedicotestE) in standard locations ofthe l2-Iead ECG sptem. An additional two electrodeswere applied to the chest for simultaneous ECGrecording to indicate the te.mporal reiation of

^Zsignals ö the cardiac cycle.fA period of 10 s was

measured with each configuration followed by a 4 s

reciprocal measurement where electrode sites of thecurrent injection and voltage measurement were inter-changed. Possible differences between normal andreciprocal data indicate electrode contact or otherrelaied technical inaccuracies in the measurementi241. AII data was sarnpled at a rate of 200 Hz using a12-bit analogue-to-digital converter of the CircMonand stored on hard disk for o$line processing andarralpis.

Analysis and cotnpadson of cknbal and, simulatzil ila.ta

The theoretical data from the models were comparedbetween each model and the data from the testmeasurements. The elinical data was compared be-tween the two study groups consisting of healthyvolunteers and valvular patients.

A commercial software was used for statistical analysis(STATISTICA@ for Windows 5.0 by StatSoft inc., Tuisa,USA). Simulated Zq lues were compared begwe-er-t-the

three models. Average measured' LZ dti4{awterecalculated for each configuration and subjecq averageLZ for each con-figurations was obtained from theaveraged LZ M from each healthy volunteer.Several parameters were derived from the data mea-sured from the volunteers: minimum and maximum Zq,

P. Karrypino a cl Dwcloping multiple ICG meruemens,(*itho*t 4vefa€i^t)

AZ--. /ndicating the maxirnum impedance deflectionof avglaged LZ, LZn^r the amplitr-rde of ventilationeffec/ during the recording period of each configura-tionj ratio of MnJ^7,c,, and the mean absolutepercentage error (MAPE) [25] between the individualaverage M and the average ÅZ calcuiated from all thevolunteers. MAPE describes the difference between anindividual LZ arrd the average ÅZ obtained from thestudy group with particular configuration. The para-meters were computed as average values from eachsubject for each configuration. MAPEs were calculatdfor three phases of the cardiac cycle; the wholeaveraged M, before the R-peak and after the R-peakof ECG. Pearson product-moment correlation matrixwas calculated between the derived panrmeters and thesensitivity distributions (i.e. proportional sensitivitymlues from each tissue type and tissue group)simulated by the models. In addition, the intergroup(volunteers versus patients) differences (MAPEs) werecalculated for the averaged

^ZFtrret

Results

Sinulakn sen sitiviE dbtributitrLs

Figure I summarizes the simulated measurementsensitivities of the configurations discussed above.Values are indicated for each tissue type in additionto three tissue groups consisting of puknonary circula-tion. wstemic circulation and all the blood masses andthe heartmuscle. The agreementbetween the differentmodels was notable and the rnlues in figu.re I areaverzrges as calculated from the three models.

Basal impednues

The simulated rzlues of Ze showed statistically signifi-cant correlation between the mod-els (p<0.008). For

the selected configurations the values were on theaverage 1.39, 1.31 and 1.33 Q for the EDM, ESM andVHM, respectively. The measured basal impedancesfrom the volunteers deviated from the sirnulated 0.79 Oon the average with significant correlation only to theVHM (p<0.05). Excluding configurations wi*r mea-sured 26 remaining zero (indicating problems with theinsmrmentation measuring small Zs), 89% (141/758)of the simulated lalues of Ze fell beween the minimumand maxinum nalue measured from the volunteers.

Cmrelatim data

The calculated correlation matrix showed a largenumber of statistically significant relationships beweön*1s 5imrrleted data and the parameters derived from themeasurements. However, the strongest value for corre-lation was only 0.45 as calculated between the positivearea after the R-peak and the sensitivity proportion inright ventride in ESM (p<0.000) and also for theMAPE after the Rpeak and the Zs change betveen theECG triggered models (p<0.000).

^7*r did not show

correlation to Zo simulated with any of the models, butsignificant correlation was found to *re ÅZ calculatedby subtracting valua of Zo calculated wittr the EDM andESM (r=0.30, p<0.000). Proportional sensitivitychange as calculated with the EDM and ESM correlatedpositively with AZ*. when investigating the heartregion (r=0.31, ?<0.000), while negative correlationwas found for the sensitivity in the lungs, the VHMshowing tlre strongest correlation (r= -0.43, p<0.000).The amplitude of respiratory deflection ÅZ*., showedpositive correlation with the ESM and EDM Zo Q=0.23...0.24, p<0.000) and sirnilar values witå the sensitivityin the group of skeletal muscle and åt tissue. Negativecorrelation was found for several tissues and tissuegroups of the cardiovascular qntem. Negative correla-tion was noted between the Ä2.,*, and the sensitivity inthe lungs with the WIM (r= -0.29, p<0.000). Ratio of

åo(t

äaooeo2oeGE810

Eått@tdc LdlEdd. !.Abg S&dåtMud! Bd&Blood SYsdrrf ldffrie AsEb D*ryb hfEffi hlEvc,!Edtbt luilt6rtfdr RSlthD8 Fd Plbctlct Iö!Lt|!i@ Addcr.å SqvaqhPEhorbry O66blmd

Figure 1 Simalautl ualues fm sensitivirizs in difnar.t tisxm of thc conf.gurations sebctzd fo clinital acpoiments. Valuzs arecabulatzd as aaerags ftnn tfu thru diffnmt fuM mailels. Tlu catnal tznil"enq of thz smsitivily in tzrrns oJ tlw ,rud,ian of the uahusis rQresmted b1 tlu nnllzst box in tlu Ph\ thz sprnd. (aariabilit| 4 tht gartilzs Qhz 25th and 75th percenti.les, lngr box in th.e

plot) ann fiu tninimurn and maimutn aahus oJ thz sensitilliry indicating thz highzst and Imtest smsitiztity amang thzcutfgurations.

P. Karppino d d! Developing multiple ICG mruuements

LZ/LZ\".. indicated strongest positive correl.iation tosensitivity in right ventricie (r=0.36, 1<0.000) andnegative to the fat tissue (r= -0.26, 1<0.000). Thesensitivity in the lungs correlated negatively with the\rlIM (r= -0.I4, p<0.027).

Intcr-individnnl a ariations zuithin uoluntzas

Generalln increasing MAPEs correlated positively with*rc h difference between the triggered models(r=0.24...0.44, p<0.000), tle sensitivity change intlre åt and skeletal muscle (r=0.24,...0.43, p<0.000)and also the heart region r=0.29...0.37, p<0.000).Negative correlations (i.e. smaller MAPE wittt highersensitivity) were evident for lung configurations witl allof the three models (r= -0.23... -0.32, P<0.000), butnot for the sensitivity change obtained by subtractingthe triggered model values, and for the tissue groupsconsisting of either pulmonary or s'6temic circulation.trigure 2 shows averaged ÅZ lalues from the volunteersfor the configurations producing the smallest (27%)and the largest (450%) inter-individual MAPEs for theperiod ofthe whole cardiac cycle.

Difnmces betueen aoluntzen and patients

The intergroup MAPEs betr,rreen the volunteers and thevalvular patients showed the highest positive correla-tions to the sensitivity in VHM heart muscle (r=0.36,

?<0.000). MAPE before the R-peak correlated posi-tively to the sensitivity change between the triggeredmodels in the atria (r=0.38, P<0.000), and after the R-peak to the heart muscle and left venricle (r=0.20,p<0.003). The largest negative correlation was foundfor the sensitivity in the lungs in the trigered models

(r= -0.25, p<0.000). Example averaged ÄZs recordedfrom both of the study populations are shown in 6gure3. For several configurations, the differences betweenthe study groups and the individuals were comprehen-sive in the morphology and timing of deflections in Å7.Certain configurations recorded similar waveformsbetr,veen the snrdy groups (a), several produced similarshapes in the impedance deflection but shifting thetime instant of t}re peak change (b). The maximumimpedance deflection was more often delayed thanadvanced with the patients when compared to thehealthy subjects although the patiens heart rate washigher (average Hk volunteers 62, patients 7l). Figure3c shows configurations with additional or missingdeflections between the study groups, and figure 3ddata with larger deviations between the groups.Investigating LZ ,frM from individuals reveals atendency shown iry'the avenged signals, although some

/6;3^*L5

MAPE II5O %

Ed^(hA^*=€c^_

o 0.5 1-0 8 0 0.5 '1.0 t

Figre 2 A.Z wouefonns obtairud tuith tzao difamt confg-urations *hibiling the smallzst and. l.a,rgat aorintirns (MAPE)betuEen tlu aohtntzen. Thin hnx shau i,nd.ittilunl waagedLZ thi.ha tracing tlu aa*age frorn all tlu ulunteets.

^{V\ t"d\

Figne 3 Exa,rnplzs oJ tlu 12-bad baseil ICG recordings shoum as araage signak fron thz studg groups (uolunters and, aahrularpaticttts). (a) Tracings tuith small intcr€roup variation (b) changa beaum tlu groups in tlu tirru instant of thz marimumimpedance d4lectbn. (c) charactaistit sign4k with rutnblz peahs or deflntiarc missing beween thz grmps. (d) Iarge inter-grouPMAPB. Fm aplnnation of tracings mar*al i and ii, plzase refer to td.

Ldry

MM

/tfr

AV\A'ry\A| ,,1 \

l-iv+ri

V-il.i ',*I'r;\'-

AU\Arll \

'AL'Y L

år'v

MAPE?7 %

ro,i01.i*,

-^\--ii

P. Kauppino et ol Dseloping multiple ICG mcruements

/ moleconfiguråtions seem to detect'regional information Ådlthus producing more diversified impedance curves. Asan example, in figure 3c the distinct peak marked ioccurs in the data recorded from 71/72 volunteers, butnot in any patienl The peak marked ii, on the otherhand, is elevated n 6/9 patients and in only onevolunteer.

Discussion

Shrtulntun sensitiviA distributittrs showed, enhanced. conhibu-taons

The simulation results obtained imply that it is notpossible to separate exdusively the contribution fromany single organ or blood mass from the measurementusing the ICG configurations based on the l2-lead ECGelectrode qÄtem. The simulated proportional contribu-tions from the classified tissues of the cardiovasculars)stem remained less than 30Vo w}l.en calculated as anaverage of the three thorax models (figure 1). Forsingle classified blood masses in the model, e.g. largeblood vessels, less than l07o contributions wereobtained. A markedly increased value of 75Vo was notedfor the group of tissues consisting of cardiovascularstuctures. The contribution from pulmonary circuia-tion was 35Vo at maximum, while from the systemiccirculation only l5%. Sensitivity along the ascendingaorta cannot be as high as in tbe pulmonary compo-nents. since elecrode locations of the l2-lead includeonly one electrode at the ankle, which was not acceptedfor simultaneous current injection and voltage mea-suremenL Using additionally the rightJeg drive elec-trode common in the l2-lead ECG recordings couldenhance the sensitivity in the aorta. However, inconventional ICGs the contributions are much lower,such as less than 77o fuom all the blood masses and theheart [9]. As compared to t]re conventional ICGs,markedly elevated sensitivities with the l2lead ICGconfigurations wslg 6ltainsfl partly as the sensitivity(scalar) field may have both positive and negative raluesreducing the total sensitivity in e.g. skeletal musde toaltnost zero.

Statistbal4 signif.cant correlztio betuesn simulatioru andmzasurernm8

Preliminary clinical experimen* were conducted toobtain a first impression of the clinical applicability ofICG elecgode configurations selected from a largenumber available. The calculated correlations betweenÖe models and the measurements were generallyrather weak. A linear correlation might not describethe relationships between the sensitivity .listributjonsand the simple parameters thatwere detected from theAZ*liitiår.ia d,: noted thatwhen contribution in somepart of the thorax increases, a simultaneous decrease inother regions may actually be an important contributorpartly causing the changes in Ä2. Also, a lariety ofdifferent type of sensitivity distributions were selectedfor dinical experiments. If only certain qpe ofmeasurements had been selected, the correlatiorumight have been higher. Nevertheless, attained correla-

6

tions are indicative of the general ability of themodelling approach in developing and understandingthe propenies of various electrode configuntionsapplied in ICG.

End, slstobc and md diastoliu modzb corctitute a drymicnrdzl

The measured values of 7a were generally small,indicating the effect of regions with nega.tive sensitivity.Significant correlation was noted only to the VfIM,which is more accurate in the anatomy than thetriggered models EDM and ESM. Yet the averagedifference between tbe simulated and measured datawas less than I O. Each of the three models was static inanatomy when considered separately. Ze can beconsidered static over the cardiac cycle, while M is adynamic factor. The relations betr,yeen the maximumvalue of the dlnamic M (Lz. "*1 and simulated (static)7a values were not observed with any of the threemodels but, logicalln the difference between the ESMand EDM Zs \ralues showed significant correlation to

^Z*. As ESM and EDM were consfircted from MR

data acquired from one individual, together theyconstitute the most simple dynamic model consistingof only two phases of the cardiac cycle, yet providing661s ;salistis comparisons with the measured data.

Inng connibutian not snbstantbl

The configurations concen!-ating measurement sensi-tivity in the lungs were selected for investiga.ting apossible funrre application in pulmonary oedemadetection. Brown et aL 126l considered the lungs tobe the major origin of both the cardiac- and respiratory-related componenB in conventional ICG. In ourme*.rr.-ettL, the ÄZ* correlated positively withthe sensitivity in the heart region but negatively with thelung sensitivity. This suggests that the lungs are not amajor contributor in the measured ÅZ values, whichcan pardy be explained also by the smaller pulsatoryblood flow changes in the lungs because the pulmonaryvascular resisr'.ce is substantially lower than thesystemic resistance. ÅZo.r. decreased with increasingconcentration of the sensitivity in the cardiovascularstructures, and with the WIM also in the lungs. Regionsof the highest sensitivity in the lungs might not behomogeneously distributed corresponrling directly withthe volume changes due to ventilation, leading con-ductive pathwap and decreasing the effects ofchangingIung volume on impedance. More meaningfirl resultsabout tlre relatiorxhip between ths l\7nJM\.r.andsimulated tissue sensitivities were obtained, indicatingthat the ventricles produced high AZ* instead oflungs or for instance skeletal musde.

Heart smsitiae rwasuranentt Profune raried. infonnation

The general morphology of the recorded l2-lead ICGsignals often had additional waveforms and notches notcoinciding with those of conventional ICG signal.Waveforms very similar in morphology and timing as

reported by Patterson and Wang [27] resemblingventricular volume curve were also obtained. Inter-

P, Karppino el cl Developing multiple ICG nerucments

individual differences in ^ZÅlu€{

(MAPE) were largerin the heart sensitive measuremens and smaller in thelung and the qntemic and pulmonary circulationme:rflrrements. This lends support for the assumptionthat the configuradons somewhat measured what wasanticipated by the simulations, since changes in theiungs are generally more homogeneous and equalbetween d1g individrrals as the heart related volume.movement and flow changes.

Marhcd diffnmces in signab betuem ztohnteers dnd, patiotts

Between the study groups the di:fferences (inter-group MAPEs) were also largest with configurationsconcentrating the seusitivity in the heart region.Similarly, when the contribution from the lungs was

larger, intergroup deviations decreased. IntergroupMAPEs before öe R-peak increased most signifi-candy with the increasing sensitivity in the atria anda.fter the R-peak with the sensitivity in the ventriclesas assessed by the difference in the sensitivitiesbetween the triggered models. These obsenationsseem reasonable as in these regions the abnormal-ities related to mitral and aortic valvs .liseases

originate. Intergroup rariability with the l2-leadICG tracings was significant with certain configura-tions, as shown in figure 3b-d, but considerablesimilarities were detected with several confi.gurations(a). lnterestingly, the inst"nt of the maximumimoedance deflection could occur earlier or lateras compared to the healthy population dependingon the measurement configuration. Weak but statis-tically significant correlation was noted between thetime delay and the sensitivity of the measurement ineither left (Ieft. ventricle and aorta; positive correla-tion) or tie right side of the heart (right ventricleand pulmonary vessels; negative correladon). Thepressure is generally much lower in the pulmonarycirculation [28], which c:ur calse the shifring of themaximum impedance pealc Although the maximumsensitivities in large vessels and blood chambers wererather limited, tley were evidently enough to causechanges benueen health and disease. This impliesthat the emphasis of such measurements is differentin the regions reflecting varied ÄZ due to vah'ulardisease.

Two-phase moilel not suff.cimt fm simula,ting imped,ance curae

The theoretical models rxed in this study did notaddress the real dynamics associated with cardiacfrrnction and ventilation, as only two phases wereincluded (ESM and EDM). Wang and Patterson [29]have studied the formation of the conventional ICGwith the models derived from the same MR data as inthis study representing systole and diastole, andconclude that the efiects of geometrical changes werelarger than conductivity changes, but that a cancellingeffect occurs between the ventricular blood volumechange and the heart movement. Cancelling ofgeometrical effects was not likely to occur to thatextent with the tested l2-lead ICG configurations, thusincreasing the contribution from volumetric andgeometrical changes and enabling *re detection of ÅZ

representing these changes. Investigating more phasesof the cardiac cycle might not increase the power of t}lemethod in deriving more regional ICG configurations,since most of the configurations in clinical testing wereattained by botl the triggered models and the VIIM.More importantly, with the multi-phase modelling time-v?rying signals could be simulated for comparison withthe measured data. In addition, refinements in theresistivity lalues selected for the classified tissues couldbe made to reach better correspondence between themodel and the reality. In the present snrdy, the knowndifficulties in selecting the resistivities were not over-come, rather, the lalues were based on the literanrre[29-31]. The real values vary from subject to subject,exhibiting arrisotropic conductivity which producesvariations of imporance [32J.

I t:a a^Ls

Litnita,tions of cbnical meawrannx-/Limited recording a$e fot each configuration in-crexed the norm/ physiological variation on theaveraged LZ /dlodC, limiting the possibility of inves-tigating each configuration more thoroughly. Forinstance, effects of ventilrtion that also changeintravascular pressures did seem to induce changesin some of the measured signals, especially with thepatient population. A multi-channel ICG insmrmena-tion capable of recording multiple multilead ICGconfigurations would be practical in prospective 12-lead measuremenB reducing the overall recordingtime and allowing analyses of signals recordedsimulaneorxly. On the other hand, the purpose ofthe pilot study was to assess preliminary feasibility ofthe tested measurements. Thus. derivation of new COequation or quantification of changes due to valvulardisease are not feasible with the presented highlylimited study population, although the morphology ofsome signals had a resemblance to pulmonary flowand ventricular volume curves [33] and some hadlandmarks that at least in this population revealed theexistence of lz.lvular disease. A large divene group ofpatiens in each disease category should be investi-gated pre- and postoperatively prior to drawing anydefinite conclusions about *re l2lead ICG measure-ments. Simultaneous acquisition of haemodynamicvari:ables such as flow and pressure in differentstruchrres of the cardiovascular qntem in the angio-graphy laboratory could also increase the knowledgeof the 12-Iead method.

Conclusion

Estimation of CO from a non-inasiveiy measuredimpedance signal is an ill-posed inverse problem withno unique solution, since ÄZ is always a combinedpresentation of multiple sources. To the best of ourknowledge, anatomically detailed conductivity modelsof the human thorax were used for *re fint time inseeking ICG measurement configurations with regionalinformation contenl The resulc altrinsd demonstratethe feasibility of the method in developing andanalping ICG measwements, Recorded l2-lead ICGsignals exhibited landmarks not coinciding with those

P. Kauppino a al Doeloping multiple ICG m6ureEcnts

of conventional ICG showing resemblance to inusivedata and morphological lariations in disease. Increas-ing the mrmber of signals should increase the sensitivityof the method and the probabiJity of more accurate COestimation. In addition, configurations producingregionai inforrnation may have a wide range ofapplications apart from the essential CO estimation.However, understanding the information conveyed bythe l2-Iead ICG would require mr.rlti-phase modellingand measurements with simultaneous acquisition ofhaemodynamic variables such as flow and pressure indifferent structures of the cardiovascular sptem. Ttreinitial results described herein are prorrising, but therange of clinical relevance and potential is a matter tobe explored in future shrdies.

Acknowledgements

This work was supported financially by the Academy ofFinland, the Ragnar Granit Foundation, the MedicalResearch Fund of Tampere University Hospital and theWihuri Foundation. The authors wish to thank Profersor Robert Patterson, Minnesota University, for gener-ously supplying the image data for the truophase model.We are grateful to Pi{o Järventausta for her assistancein dinical practice and Dr Rauri khtinen for his advicein statistical anal;nes.

References

l. PATrrRsoN, R P., Kr,BrcE& W. C., Ifu.rNEr.r, E., WnsoE, D. Aand NoRxN, G., f964, Development of an elecricalimpedance plethpmography syst€m to monitor czrdiacouqrt hocel,ings of thz lst Annul Rocb Mt. BimgiwingSymposiun pp.56-71.

2. Kr,ucn& W. G., IgFNEGrs,J. N., PArrERsoN, R P., WmsoE, D..d and lvf,rrrsor, R H., 1966, Dwelopment ud evaluationof an impedance 6diac output ststem. Amspae Medicita37, 1208- 1212.

3" BERNsrrN, D. P., 1986, A new stroke volume equation forthoracic elecrical bioimpedance: theory and ntjonale.Crbical Care Medicha 14, 904-909.

4. KooBr, T., K.lr,rsxEJ.r, S., TuBenua,q, V. M. and UusrrAro, AJ., 1997, Wholebody impedance cardiography in themeasurement of cardiac otqut Crirital Cne Mdiclru- 25,779-785.

5. GEDDES, L A md BArcR, L. E., 1989, hirciplts of WUzdbionud,ial insnunmlalzon (New Yorlc John Wiley & Sons) .

6. Morupxrne" S- N., 1981, N*iwasiae catil,iansalar monilm-ing bJ ehcbbal i.mpedme t chn@u (London: Pitrnan).

7. WeNc, L. and Pantrsox, R, 1995, Multiple sources of theimpedance cardiogram based on &D finite differencehmm thorax models. IEEE Transaaiorc on Biorudicalhgtmkg 42, 141-1€.

8. P.rrrr:nsoN, R P-, WANG, L, RAZA, B. and Wooo, L, 1990,Mapping the cardiogenic impedance signd on the thoracrcsutface. Med.ital ud Biotog;tnl Zngituaing md &nputin& 28,2r2-276.

9. Ker,rpnmr-r, P. K, HrrmrnN,J..d and Meuvnvuo,J. A, f 998,Sensitivity disuibutions of impedmce cardiography usingband and spot electrodes analped by a 3D computermodel. Anmls of Bioncnial Engiwing,26, 694-702.

8

10. KAUPpTNEN, P. K, HvrrnrN,J., KööBr, T. ard Mru,wruo,J.,1999, kad field theoretical approach in bio.irnpedancemeasurements towards more confolled measurementsensitivity. Anrnk of thz Nn Yor* Aeanzny of Sciences @rPr$s). 8"3 : r35-l\e'

11. RoLFr, P., 1979, Nm-imriu physblngiral waswh(London: Academic Press).

12. LAr"BERrs, R, Vrsr,n, I(. R and ZULrm4 W. G., 1984,Inpdarce C-dbg"plry (Assen: Van Gorcum).

13. JENSEN, L., YAr<n,n.rs, J. and Tro, K K, 1995, Ä review ofimpedance cardiognphy. Heut and Lung,24, 183-193.

14. GFsElowrrz, D. 8., f97f, An application of electrocardio-graphic lead tleory to impedance plethpmography. IEEETmnsadim un Biondbal Enghzring, 18, 38 -41.

15. KAL?PD\EN, P., Hyrm{EN, J., LAARNE, P. and Meuawo, J.,1999, A softrmre implemenation for deniled volmeconductor modelling in electrophJsiology using finitediEerence method. Confuttr MethodÅ Mn hogram innAnzatu:ha 58, fgf -203.

16. Kewpwpw, P., fVrrNEN,J. and MAr-r,trwo,J., 1998, Dehiledmodel of öe thorax as a volmc conductor baed on thevisible human m datz Jowttal of Medbal Engiwing mdTuhmlog,n, n6-733.

17. MAlrtffuo, J. and Pt-oNsv, R, 1995, Bbelzdrmgndim:Pirciplzt mn appbetim of binlectric md bbmgn&b fclds(New York Offord Univenity Press).

18. I-AÄRNE, P., EsKorl" H., tlvrnNsN, J., Smxo, V. andMennwo, J., f995, Validation of a detailed computermodel for the electric fields in the brain. Jontal of Mcdialhgimzing and Tutuwbg, f9, 84-87.

f9. Hvrrwn, J., 1994, Dnelopwt of ngimul ainud ECG bn^espcizlly for ryocard.ial ischmia däagn6*;, Ph.D. disertation,Tampere University of Technologp Tampere, Finland.

20. WAr-G& S.J. and Knrernrcx, D., 1987, Fornard and inverseelecfocardiographic calculations using resistor networkmodels of the human tono. C,imJatim Reuch,61, 504-513-

21. Ampr.sN, M.J., 1991, The visible human projectJwwl ofBbmmrnhaiont18, 14-

22. HEnror.ral.r, T., DAsTDA& P., KAUIrTNEN, P., MAr-Äowo,J. andEsror-ro H., 1998, Semi-automatic tool for *gmenedonand volumetric analysis of medical images. Mediml andBiologiralEngiwns md Cmputing 36, 29f -296.

23. K^upptrrEN, P. K, Hvrrn'r,N, J. A K, Kööar, T. andMAl.rrlrvuo, J., 1999, Multiple lead recordings irnproveaccuraq/ of bioimpedance plethysmognpbic 1e-ch4i-que.Mcdiral' Engirwi"g i"d enyti;/ffi/

24. KArJppnrEN, P. K-, KööBr, T., Hyrm{EN,J. A and MAbfl'uo,Jd Assuring the quality of elecrode contacts in dinicalbioimpedance measuremens. Jfdual of Mdiral Engiwingand Tuhrobg, (submitte<i;.

25. MARTDAKS, S. G., WHEELWRTGTI|, S. C. and McGE, V. E.,L983, Forecasting mahods and appäcalaoru (New York Wiiey) .

26. BRowN, B. H., BARBER, D. C., MoRrcE, A H. and Lrerrnnn,A D., 1994, Cardiac and respiratory related elecricalimpedance changes in the human thorax. |FFFTrutsarlimm Bimdical Engircaing, a\ 729-734.

27. P.ryrr.:nsoN, R and W.rNc, L, 1998, Ventricular volmecurves obtained from thoncic impedznce measuremens.hoedings of thc 20ih Anruul Intcnatiotnl Cmfermze of therppF Enginzqing in Mcdiche orut Biobg Soeir|, Hong Kong,pp.444-446.

28- Sprnnrros, N. (ed.), 1984- Phlsiobg md pathophysiolog oftlu ]uart (Boston: Martinu Nijhotr Publishing).

29. WANG, X., Sr:lv, H. lL andVerorWert&J. M., 1995, Time-frequency disribution technique in biological signalprocessing. Binudital Ircturmmtation and, Tuhrcbg, N,203-2t2.

P. Kauppino el al Dweloping multiple ICG mwoents

30. Gf,DDEs, L- A and Beren, L. E., 1967, Tbe specific resistanceof biological material-a compendium of data for thebiomedical engineer and phpiolo$sc Mdizal and BbltgitalEngiwing ann Conpning, 3, 271-293.

31. RusH, S., ABtrJsKov,J. .{ and MiFrr, R, 1963, Resistivit}' ofbody tissues at low frequencies. Ciraidim" 22" 40-50.

32. Hrrnww, J., KewnNrN, P., KööBI, T. and Mewnruo, J.,1997, Imporance of the tissue conductivity values inmodelling the thorax as a volume conductor. hoceningsof thv 19th Annul Intzrnotioral Confamcc of tlv m?nEnghering in Mediciru and Binlng Socil1, Chicago, pp.618- 619.

Ru$$,IER, R- F., 1961, Cud.inasctnu dynonics (Philadelphia:Saunders).

Jt.