Authorized licensed use limited to: MIT Libraries. Downloaded on April - PDF document

Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 14:10:56 UTC from IEEE Xplore. Restrictions apply. 1250 IEEETRANSACTIONSONBIOMEDICALENGINEERING,VOL.57,NO.5,MAY2010 Fig.9.Long-term insitu EDArecordings.Continuousskinconductancemea- surementswererecordedforsevendaysinanaturalhomeenvironment.Daily EDAwaveformsdisplayedarenormalized. sideofthedistalforearmsarecontinuouswiththepalmarsites (C6ÐC8),butKeeganandGarrett[39]describetheventralside ofthedistalforearmsasinnervatedbydermatomesC5,C6,C8, andT1.Inaddition,differencesinmoisturebuildupbetween theskinÐelectrodeinterfacescouldalsocontributetothelarge varianceofcorrelationcoefÞcients.Thiswasparticularlyevi- dentattheendofthephysicaltask,whichresultedinnegative- correlationcoefÞcientsin38%ofparticipants(decreasingskin conductanceatÞngers,butincreasingskinconductanceatdistal forearms).Thereisalackofevidencetosuggestadependence onageorgender.Nonetheless,theoverallpatternofresultssug- geststhattheEDAmeasuredfromthedistalforearmsclosely parallelsEDAmeasuredfromtheÞngers. Usingconductivefabricaselectrodesisanattractiveoption indesigningwearablesensorsasitpotentiallyenablesgreater comfort.Althoughatrestandrecovery,theconductivefab- ricelectrodesperformedsomewhatsimilarlytoconventional Ag/AgClelectrodes,itisimportanttotakenotethatthatthey haveweakerabilitytomeasureEDAchangesduringstressor tasks.Giventhestretchynatureoftheconductivefabric,itis likelythattheelectrodesdonotmaintaintheirelectricalprop- ertieswhenincontactwiththeuser.Furthermore,theabilityof fabrictoabsorbmoistureandsweatalsocontributestoaltering theirelectricalpropertiesovertimeandisthelikelyreasonfor ourresultsshowingthatthelevelsofEDAwerehigherafterthe physicalexertiontaskfortheconductivefabricelectrodesthan forthetraditionalAg/AgClelectrodes.Inviewofthis,werec- ommendusingstandardAg/AgClelectrodesformoresensitive analysisofEDAmeasurements. Long-termassessmentofEDArevealedinterestingtrends intheparticipantÕssympatheticmodulationoveraweek-long period.IntervalsofelevatedEDAfrequentlycorrespondedto timeswhentheparticipantwasstudying,doinghomework,or takinganexam.Thisispossiblyduetotheincreasedcognitive stressassociatedwiththeseactivities.Thecharacteristicpeaks occurringduringsleephavebeenassociatedwithslow-wave sleep[40]andremainasubjectforfuturestudies.Wefound thesensorwristbandtoprovidereliableandrobustattachment oftheelectrodestotheskin,eveninthepresenceofforearm motionduringnormaldailyactivities.Motionartifactswere typicallyobservedonlywhentheelectrodeÐskininterfacewas disturbedsuchaswhenexternalpressurewasappliedagainst theelectrodesorwhenthewearerreadjustedthepositionofthe electrodes. VI.C ONCLUSION Wehavepresentedacompactandlow-costwearableEDA sensorthatenablescomfortablelong-termassessmentofEDA. Thenoveltyofoursystemconsistsoftheuseofthedorsal forearmsasrecordingsites,theminiaturizationofthesensor module,aswellasthedesignofasmallwristbandthatallows forunobtrusiveandnonstigmatizingcontinuousEDAmeasure- mentsduringeverydayactivities.Experimentaloutcomesusing Ag/AgClelectrodescorrelatedstronglywiththeFDA-approved EDAmeasurementsystem.Tothebestofourknowledge,we describedtheÞrstdetailedstudyindicatingthattheventralside ofthedistalforearmsisaviablealternativetothemorepopular palmarsitesforEDAmeasurementsacrossphysical,cognitive, andemotionalstressors.Importantly,wealsopresentedtheÞrst long-termrecordingsofEDAduringdailyactivityoutsideof alaboratoryorclinicalsetting.Whilepalmarelectrodesare encumbering,easilylost,andfrequentlysubjectedtomotion andpressureartifacts,theproposedwrist-wornsensordoesnot sufferanywherenearasmuchfromtheseproblems.Giventhe versatilityoftheproposedsystemthatactsbothasadatafor- wardinganddataloggingdevice,usersarenotconstrainedto staywithintherangeofabasestation,but,instead,haveun- restrictedcontinuousmeasurementsregardlessoflocation.The importanceofthispaperistheunprecedentedabilitytoperform comfortablelong-termand insitu assessmentofEDAthatthe proposedsystemoffers.Long-termcontinuousEDAmeasure- mentsduringnormaldailyactivitylikethatinFig.9have,tothe bestofourknowledge,notpriortothispaperbeendemonstrated inapracticalway,andthus,thenewtechnologydevelopedin thispaperrepresentsasigniÞcantadvancementoverexisting systems.Investigationsoflong-termsympatheticnervoussys- temactivitycanpotentiallyaddpreciousinsightandenrich understandingofwidespreadneurologicalconditions.Studies arecurrentlyunderwaytoevaluatetheuseoftheproposedEDA sensorinavarietyofclinicalapplications,includingautism, epilepsy,andsleepdisorders. 1246 IEEETRANSACTIONSONBIOMEDICALENGINEERING,VOL.57,NO.5,MAY2010 (0801002576)wasapprovedbytheMassachusettsInstituteof TechnologyCommitteeOntheUseofHumansasExperimen- talSubjects(COUHES).Informedconsentwasobtainedfrom allparticipantspriortothebeginningoftheresearchsession. Sixteenparticipants(eightfemalesandeightmales)wereen- rolledinaphysicaltask,15participants(ninefemalesandsix males)underwentacognitivetask,and13participants(eightfe- malesandÞvemales)weresubjectedtoanemotionaltask.Due tobatteryfailureontheFlexcomp(oneperson)andmotion- corrupteddata(oneperson),full-lengthrecordingswerenot availablefortwoparticipantsduringthecognitivetask,butun- corruptedsectionswereincludedinanalysis.Weexcludeddata fromthreeparticipantsinthephysicaltaskduetofailuretoturn theproposedsensoron(oneperson),disconnectionofÞnger electrodesfromtheFlexcompduringtask(oneperson),and motion-corruptedÞngerdata(oneperson). Inaddition,oneparticipantvolunteeredtoweartheproposed sensorforalong-termexperimenttomeasureEDA insitu . B.PhysiologicalMeasurements EDAwasmeasuredasskinconductancechangesusingei- thertheproposedsystemoraÒgoldstandard,Ócommercially availableandFDA-approveddevice(FlexcompInÞniti,Thought Technologies,Ltd.)atfourdifferentrecordingsites(seeFig.4). Ag/AgClelectrodesembeddedinhook-and-loopfastenerbands weresecuredaroundthemedialphalangesoftheindexand middleÞngersbilaterally.TheelectrodesontherightÞngers wereconnectedtotheFlexcompInÞniti(whichservedasour control),whiletheelectrodesontheleftÞngerswereconnected totheproposedsensormoduleforEDAmeasurements.Awear- ableEDAsensor(wristband)wasplacedoneachdistalforearm withthesensormoduleontherightconnectedtoAg/AgClelec- trodes,andthesensormoduleontheleftconnectedtostretch conductivefabric(silverplated92%,nylon8%,surfaceresis- tance 1
/sq,andcontactareaof3.5cm 2 )thatwassewninto thewristbandintheplaceoftheAg/AgClelectrodes.Electrodes forbothwristbandswereincontactwiththeventralsideofthe distalforearms.Therewasnopretreatmentofrecordingsites andnoconductivegelwasappliedtotheelectrodes.Theclocks fortheproposedsensorsandtheFlexcompweresynchronized priortothestartofeachexperiment,andthetimewasrecorded atthebeginningandendofeachcondition.Thesamplingfre- quencyforallsignalswasÞxedat32Hz. C.PhysicalTask Duringthebaselineperiod,participantswereaskedtosit quietlyontheseatofarecumbentbicycle(PrecorUSA,Inc.) andrelaxfor10min.Participantswerethenaskedtopedalas fastastheycouldforadurationof5minataÞxedresistanceof level7.Afterthecyclingtask,participantswereaskedtoremain seatedforarecoveryperiodof10min. D.CognitiveTask Participantswereseatedinadarkenedandsound-dampened roomfacingalargescreenandbrießyorientedtotheirsurround- ings.Atthestartoftheexperiment,participantssatquietlyfor 10mintoobtainabaselinerecording.Participantswerethen askedtofollowinstructionsonthescreenforacognitivetask. On-screeninstructionsrequestedthemtoperformserialsub- tractionsinintervalsofseven,startingwithafour-digitnumber, andtodelivertheirresponsesvocallyfora3-minperiod.A countdowntimerwasvisibletotheparticipantsandabuzzer wassoundedforeachmistakeinsubtraction.Thementalarith- metictaskwasfollowedbyaStroopwordÐcolormatchingtest. Aslideconsistingofcolorwords(red,blue,green,brown,and purple),eachprintedinacolordifferingfromthecoloritnamed wasdisplayedonthescreenandparticipantswereaskedtoread thenamesofthecolorsseriallyasquicklyaspossiblein1min. Onceagain,acountdowntimerwasvisibleandparticipantswere alertedtoerrorsbyabuzzer.Attheendof1min,thetaskwas repeatedagainwithasecondslidethathadthewordsprinted inthereverseorder.Aftercompletionofthetask,participants wereaskedtoremainseatedandrelaxforarecoveryperiodthat lastedfor10min. E.EmotionalTask Similartoprevioustask,participantswereseatedinadark- enedandsound-dampenedroomfacingalargescreenandbrießy orientedtotheirsurroundings.Atthestartoftheexperiment, participantssatquietlyfor10mintoobtainabaselinerecording. Afterestablishingabaseline,participantswereinformedbyan investigatorthatahorrormoviewouldbeplayedandwaitedin anticipationfor1min.A5-minclipfromthemovieÒ28Days LaterÓinvolvingscenesofchaos,violence,anddisturbingim- ageswasthenshown.Attheendofthemovieclip,participants wereaskedtoremainseatedandrelaxforarecoveryperiodthat lastedfor10min. F.Long-TermInSituExperiment Ahealthyvolunteer(19-year-oldmale)wasinstructedtowear theproposedsensorwithAg/AgClelectrodesonhisleftdistal forearm24/7foraweektomeasurelong-termEDAduring dailyactivities.Hewasgiventheoptiontostopparticipating intheexperimentatanymomentandtoremovethewristband atanytimehechose.Afteraperiodbetween24and30h,the volunteerremovedthemicro-SDcardtodownloadthedataand alsoreplacedthebattery.Thiswasrepeateddailyforatotalof sevendays. G.DataAnalysis AlldataÞleswereanalyzedusingcustomsoftwarewritten inMATLAB(TheMathWorks,Inc.).TherawEDAsignals wereÞlteredwitha1024-pointlow-passÞlter(Hammingwin- dow,cutofffrequencyof3Hz)toreducemotionartifactsand electricalnoise.PearsonÕscorrelationcoefÞcientsandthecor- responding p -valueswerecalculatedfortheÞlteredrecordings fromthedifferentsitesandsystemsasameasureofsimilarity betweensignals. 1252 IEEETRANSACTIONSONBIOMEDICALENGINEERING,VOL.57,NO.5,MAY2010 [27]C.Tronstad,G.E.Gjein,S.Grimnes,O.G.Martinsen,A.L.Krogstad, andE.Fosse,ÒElectricalmeasurementofsweatactivity,Ó Physiol.Meas. , vol.29,pp.S407ÐS415,Jun.2008. [28]M.Strauss,C.Reynolds,S.Hughes,K.Park,G.McDarby,andR.W. Picard,ÒTheHandWavebluetoothskinconductancesensor,Ópresentedat the1stInt.Conf.AffectiveComput.Intell.Interact.,Beijing,China,2005. [29]Y.B.Lee,S.W.Yoon,C.K.Lee,andM.H.Lee,ÒWearableEDAsensor glovesusingconductingfabricandembeddedsystem,Óin Proc.Conf. Proc.IEEEEng.Med.Biol.Soc. ,2006,pp.6785Ð6788. [30]F.Axisa,C.Gehin,G.Delhomme,C.Collet,O.Robin,andA.Dittmar, ÒWristambulatorymonitoringsystemandsmartgloveforrealtimeemo- tional,sensorialandphysiologicalanalysis,Óin Proc.26thAnnu.Int.Conf. IEEEEng.Med.Biol.Soc. ,SanFrancisco,CA,2004,pp.2161Ð2164. [31]W.H.Rickles,Jr.,andJ.L.Day,ÒElectrodermalactivityinnon-palmar skinsites,Ó Psychophysiology ,vol.4,pp.421Ð435,Apr.1968. [32]D.Shastri,A.Merla,P.Tsiamyrtzis,andI.Pavlidis,ÒImagingfacialsigns ofneurophysiologicalresponses,Ó IEEETrans.Biomed.Eng. ,vol.56, no.2,pp.477Ð484,Feb.2009. [33]D.C.Fowles,M.J.Christie,R.Edelberg,W.W.Grings,D.T.Lykken, andP.H.Venables,ÒCommitteereport.Publicationrecommendationsfor electrodermalmeasurements,Ó Psychophysiology ,vol.18,pp.232Ð239, May1981. [34]S.Fisher,ÒBodyimageandasymmetryofbodyreactivity,Ó J.Abnorm. Psychol. ,vol.57,pp.292Ð298,Nov.1958. [35]P.A.Obrist,ÒSkinresistancelevelsandgalvanicskinresponse:Unilateral differences,Ó Science ,vol.139,pp.227Ð228,Jan.18,1963. [36]Y.Kuno, HumanPerspiration .Oxford,U.K.:BlackwellScientiÞcPub- lications,1956. [37]J.A.Allen,J.E.Armstrong,andI.C.Roddie,ÒTheregionaldistribution ofemotionalsweatinginman,Ó J.Physiol. ,vol.235,pp.749Ð759,Dec. 1973. [38]F.A.Fender,ÒFoersterÕsschemeofthedermatomes,Ó Arch.Neurol. Psychiatry ,vol.41,pp.688Ð696,1939. [39]J.J.KeeganandF.D.Garrett,ÒThesegmentaldistributionofthecutaneous nervesinthelimbsofman,Ó Anat.Rec. ,vol.102,pp.409Ð437,Dec.1948. [40]L.C.JohnsonandA.Lubin,ÒSpontaneouselectrodermalactivityduring wakingandsleeping,Ó Psychophysiology ,vol.3,pp.8Ð17,Jul.1966. [41]R.Edelberg,T.Greiner,andN.R.Burch,ÒSomemembraneproperties oftheeffectorinthegalvanicskinresponse,Ó J.Appl.Physiol. ,vol.15, pp.691Ð696,Jul.1960. Ming-ZherPoh (SÕ09)receivedtheB.Sc.(Hons.) degree( magnacumlaude )inelectricalandcomputer engineeringin2005fromCornellUniversity,Ithaca, NY,andtheS.M.degreeinelectricalengineeringin 2007fromtheMassachusettsInstituteofTechnology (MIT),Cambridge,whereheiscurrentlyworking towardthePh.D.degreeinelectricalandmedical engineeringattheHarvard-MITDivisionofHealth SciencesandTechnology(HST),Cambridge. HewaswiththeMassachusettsGeneralHospital (MGH)CenterforEngineeringinMedicine,where hewasengagedindevelopingmicroßuidicbiochips,andwaswiththeMGHDe- partmentofRadiationOncology,wherehewasinvolvedinoptimizingdelivery ofnanoparticlestotumors.Since2008,hehasbeenwiththeAffectiveCom- putingGroup,MITMediaLaboratory.Hiscurrentresearchinterestsinclude neurophysiology,wearablebiosensors,andassistivetechnologiesforneurolog- icaldisorderssuchasepilepsyandautism. Mr.PohisalsoaStudentMemberoftheTauBetaPiandanAssociate MemberoftheSigmaXi. NicholasC.Swenson isworkingtowardtheB.Sc.de- greeinbiomedicalengineeringattheMassachusetts InstituteofTechnology(MIT),Cambridge. HeiscurrentlyamemberoftheAffectiveCom- putingGroup,MITMediaLaboratory.From2006to 2007,hewaswiththeRiversideHospital,Newport News,VA,wherehewasengagedinresearchusing hyperbarictherapyonwoundsindiabeticpatients. From2007to2008,hewaswiththeMaryImmac- ulateHospital,Grafton,VA,wherehewasinvolved inresearchonthevenousbloodßowduringdifferent total-hiparthroplasties. RosalindW.Picard (MÕ81ÐSMÕ00ÐFÕ05)received theBachelorsdegree(withhighesthonors)inelec- tricalengineeringfromGeorgiaInstituteofTechnol- ogy,Atlanta,in1984,andtheS.M.andSc.D.degrees inelectricalengineeringandcomputersciencefrom theMassachusettsInstituteofTechnology(MIT), Cambridge,in1986and1991,respectively. From1984to1987,shewasamemberoftheTech- nicalStaffatAT&TBellLaboratories,Holmdel,NJ. ShewasanInternatHewlettPackard,IBM,andSci- entiÞcAtlanta.ShewasaConsultantatavarietyof companies,includingApple,IRobot,BT,andMotorola.SheiscurrentlyaPro- fessorofmediaartsandsciencesattheMITMediaLaboratory,wheresheis alsotheFounderandtheDirectoroftheAffectiveComputingGroup,andthe LeaderofanewAutismandCommunicationTechnologyInitiative.Sheisalso Co-founder,Chairman,andChiefScientistofAffectiva,Inc.Shehasauthored orcoauthoredmorethan200scientiÞcarticles.Sheisbestknownforpioneer- ingworkinimageandvideocontent-basedretrieval(theoriginalPhotobook system),fordevelopingtexturemodelsandmachinelearningfortheircombi- nation(SocietyofModels),andforherbook AffectiveComputing (MITPress, 1997),whichhelpedlaunchaÞeldbythatname.Hercurrentresearchinterests includethedevelopmentoftechnologytohelppeoplecomfortablyandrespect- fullymeasureandcommunicateaffectiveinformation,andthedevelopmentof modelsofaffectthatimprovedecisionmakingandlearning. Prof.PicardisamemberoftheAssociationforComputingMachinery.She wasarecipientofaBestPaperPrizefromtheIEEEInternationalConference onAdvancedLearningTechnologiesin2001. POH etal. :WEARABLESENSORFORUNOBTRUSIVE,LONG-TERMASSESSMENTOFELECTRODERMALACTIVITY1249 Fig.8.DistributionsofcorrelationcoefÞcientsbetweenEDAmeasurementsfromdistalforearms(conductivefabricelectrodesontheleftdistalfo rearmand Ag/AgClelectrodesontherightdistalforearm)andipsilateralÞngersunder(a)physical( n = 13),(b)cognitive( n = 15),and(c)emotional( n = 13)stressors. Eachexperimentwasseparatedintoabaseline,task,andrecoveryconditionsforcorrelation.ThemedianvaluesofthecorrelationcoefÞcients  r alongwiththe performanceindexes(proportionofcorrelationcoefÞcients  0.5)
arealsopresentedforeachcondition.Dashedlinesindicate r = 0.5.(a)Physicalstressors. (b)Cognitivestressors.(c)Emotionalstressors. habituation,whichischaracterizedbydecreasingreactionin- tensitywithrepeatedstimulation,performingrepeatedmeasure- mentswiththedifferentsensorsisalsonotapracticalsolution. Thisproblemcanbeavoidedbyplacingthedifferentpairsof electrodesonbilateralrecordingsites,buttherestillexistsissues concerningleft-hand/right-handdifferencesinEDA[34],[35]. Thus,itisimportanttopointoutthefactthatevenwithidentical sensors(e.g.,Flexcomp)placedonbilateralsites,oneshouldnot expecttoobtainidenticalmeasurements.Withthiscaveat,we formulatedtheexperimentalsetupdescribedinSectionIII-Band depictedinFig.4withthegoalofcomparingtheperformance characteristicsoftheproposedEDAsensorwithaÒgoldstan- dard,ÓFDA-approvedandcommerciallyavailableEDAmea- surementsystem(Flexcomp). Thethreeexperimentsperformedwereselectedtocapturea varietyofEDAresponsestoclassicstimuli.Furthermore,the cognitiveexperimentconsistedoftwoseparatetasksofdiffering complexity(mentalarithmetictaskandtheStroopwordÐcolor matchingtest),whiletheemotionalexperimentalsoelicitedan anticipatoryresponsepriortoprovidingemotionalstimuli.It iswellknownthatstresscaninduceanincreaseinbothtonic andphasiccomponentsofEDA.OurmeasuredEDArecordings showedchangesofskinconductancefrombaseline(atrest)in responsetoastressortaskandduringrecovery.Asexpected, skinconductanceincreasedduringphysicalstrain,inducedby intenseexercise,andemotionalstrain,inducedbymentalarith- metic,theStroopwordÐcolormatchingtask,andthehorror movieclip.Althoughitistemptingtoassumethateachdiffer- entEDAwaveformrepresentseitherpurelythermoregulatory oremotionalsweating,itisworthnotingthattheresponsesare likelyacombinationofdifferentsweatingmechanisms.Forex- ample,sweatingfromphysicalexerciseinvolvesboththermal andmentalsweating[36],andtherapidchangefrombaseline atthestartofthevariousstressortaskspresumablyinvolvesthe orientingresponse.Therefore,theserecordingsmerelydemon- stratechangesinEDAunderdifferentstressors.Emotional sweating(increasedsweatglandactivityasaconcomitantof psychological,andespecially,emotionalstates)hasbeenob- servedmainlyonpalmarandplantarsites,butthespeciÞcityof emotionalsweatingremainsinquestion[37].OurÞndingsindi- catethattheventralsidesofthedistalforearmscanalsoproduce sweatresponsestoemotionalstimuli.Thestrongcorrelations betweentheproposeddeviceandtheFlexcompindicatethatour systemfunctionsasintended. ThewidedistributionincorrelationcoefÞcientsbetweenthe Þngersandthedorsalforearmcouldbeduetodifferencesin sweatglanddistributionorskinsudomotorinnervation.Indeed, thetwomostcommonlyuseddermatomemapspresentdiffer- entspinalnerveinnervationsattheventralsideofthedistal forearms.AccordingtoFender[38],dermatomesoftheventral 1244 IEEETRANSACTIONSONBIOMEDICALENGINEERING,VOL.57,NO.5,MAY2010 sympatheticarousalinautism[18],earlydiagnosisofdiabetic neuropathy[19]andprovidingbiofeedbackintreatingchronic hyperhidrosis[20],epileptic[21],andpsychogenicnonepilep- tic[22]seizures. Atpresent,ambulatoryEDAdevicesareoftencomposedof aprocessingunit/AÐDconverterandexternalprobesthatre- ducecomfortlevels.CommercialsensorsystemssuchasFlex- Comp[23],SenseWear[24],QPET[25],andVitaport[26]offer high-qualityEDArecordings,buttheircurrentbulkyformfac- torsandhighpriceslimittheirwidespreaduseforlong-term ambulatorystudies.Similarly,Tronstad etal. [27]proposeda portableloggerforEDAlong-termmeasurements,butitssizeis stillconsiderablytoolarge(157mm × 95mm × 33mm)tobe practicalforcontinuousmonitoringortobeconsideredawear- abledevice.TheHandWaveBluetoothskinconductancesensor representsasigniÞcantimprovementintermsofsize[28].How- ever,likemanyothersimilardevices[29],[30],theplacement ofelectrodesontheÞngersorpalmsrequiresexternalconnec- tions,andisencumberingandhighlysusceptibletomotionor pressureartifacts[31].ArecentproposaltomeasureEDAun- obtrusivelythroughimagingmeans[32]offersmuchpromise, butthemeasurementscarrysubstantialnoiseandquantiÞcation remainsdifÞcult. Toachievewidespread,continuous,andlong-termassess- mentofEDA,thereisaneedforasensorthatnotonlyislow cost,compact,andunobtrusive,butalsocomfortabletowearand nonstigmatizingtotheuser.Inthispaper,wepresentanovel solutionintheformofawearableandfullyintegratedEDA sensorthatfulÞllsthesecharacteristics.Thestudyfocuseson comparingtheperformanceoftheproposedsystemwithaFood andDrugAdministration(FDA)approvedEDAmeasurement systemduringclassicarousalexperimentsinvolvingphysical, cognitive,andemotionalstressors.WeÞrstvalidatetheperfor- manceoftheproposedsensorduringEDAmeasurementsfrom traditionalpalmarrecordingsites.Inaddition,westudytheuse oftheventralsideofthedistalforearmsasarecordingsitefor EDAmeasurementsthatisnonencumbering.Wealsoinvesti- gatehowthechoiceofelectrodematerialaffectsperformance bycomparingtheuseofconductivefabricelectrodestostandard Ag/AgClelectrodes.Finally,wepresentaweek-longrecording ofEDAduringdailyactivity.Tothebestofourknowledge,it istheÞrstdemonstrationoflong-term,continuousEDAassess- mentoutsideofalaboratorysetting. II.S YSTEM D ESIGN A.SensorModule TheoverallsystemisillustratedinFig.1.Direct-current(dc) isappliedtothestratumcorneumbeneathmeasuringelectrodes forexosomaticmeasurementsofEDA.Toachieveawidedy- namicrangeofskinconductancemeasurements,theanalogcon- ditioningcircuitryutilizesnonlinearfeedbackautomaticbias controlwithlow-poweroperationalampliÞers(LTC6081by LinearTechnology).Atriple-axisaccelerometer(ADXL330by AnalogDevices,Inc.)isalsoincludedforphysicalactivitymea- surements.Accelerometryisalow-cost,ßexible,andaccurate methodfortheanalysisofpostureandmovement. Fig.1.OverviewoftheEDAsensorsystemarchitecture.Thedeviceiscapable ofrecordingmeasurementsontoanonboardßashmemorycard(datalogging), wirelesslytransmittingdatatoaremotesite(dataforwarding),andperforming real-timeanalysis(dataprocessing). Adigitalsignalcontroller(DSC)(dsPIC30F2012byMi- crochipTechnology,Inc.)actsasthecontrolcenterthatcan beprogrammedonboardthroughanin-circuitserialprogram- ming(ICSP)interface.DSCscombinethecontrolattributesofa microcontroller(MCU)andcomputationcapabilitiesofadigital signalprocessor(DSP),thusallowingapplication-speciÞcreal- timecomplexanalysisonboard.Theanalogsignalsaresampled at32HzviaanAÐDwith12-bitresolutionontheDSC.Power isdrawnfromasinglelithiumpolymerbatterywithanominal voltageof3.7Vandacapacityof1100mAh.Thebatterycanbe rechargeddirectlyfromauniversalserialbus(USB)portbyan onboardsingle-cellLi-ionbatterycharger(LTC4062byLinear Technology).Astep-up/step-downchargepump(LTC3240by LinearTechnology)producesaÞxed,regulatedoutputof3.3V fortheDSCandperipheralcomponents. InordertoenablecontinuousmeasurementsofEDAand physicalactivitywithouttheconstraintofstayingwithinrange ofabasestation,adataloggingsystemisavailableonboard. UsingaseparateMCUwithdedicatedÞrmwaretoimplement aFAT32Þlesystem(uALFATbyGHIElectronics,LLC)that communicateswiththeDSCthroughauniversalasynchronous receiver/transmitter(UART)interface,datacanbewrittento removableßashmemorycard.A2GmicroSDcardprovides enoughstoragecapacityforupto28daysofcontinuousmea- surementswithasamplingrateof32Hz.Ifitisdesirableforthe datatobeaccessibletothewearerÕscaregiverforanalysisandin- terpretation,orifthewearerchoosestosharehis/herrecordings, theproposedsystemcanalsooperateasadata-forwardingde- vicewiththeuseofa2.4-GHztransceivermodule(nRF2401A bySparkfunElectronics).Inthismode,real-timemeasurements canbedisplayedonaPCequippedwithaseparatetransceiver POH etal. :WEARABLESENSORFORUNOBTRUSIVE,LONG-TERMASSESSMENTOFELECTRODERMALACTIVITY1245 Fig.2.EDAsensormodule.Thedevicehasamodulardesignandisshown withanoptionalradiotransceivermountedontop. Fig.3.WearableEDAsensor.(a)Finalpackaginginanattractiveandincon- spicuouswristband.(b)DisposableAg/AgClelectrodesattachedtotheunder- sideofthewristband.(c)WearableEDAsensorcanbeworncomfortablyon thewristforlongperiodsoftimeandduringdailyactivities. moduleforimmediateanalysis.Thecompleteelectronicmodule (20mm × 30mm × 0.8mm)isshowninFig.2. B.Packaging Usercomfortisamajorconsiderationinthedesignofany wearabledeviceintendedforlong-termandcontinuoususe.Re- gardlessofthecapabilitiesofawearablesystem,userswillnot beinclinedtowearthemonadailybasisoveraperiodofdaysor weeksifthesensorsarebulkyandcumbersome.Inviewofthis, weintegratedtheelectronicmoduleintoaregularwristband madeoutofterrycloth,resultinginacomfortable,attractive, andlightweightwearablesensor(seeFig.3).Sinceallelectron- icsandwiringareconcealedwithinthewristband,theresulting Fig.4.Experimentalsetup.Measurementswererecordedfrom(I)rightÞngers withtheFlexcompsystem,(II)leftÞngerswiththeproposedsensormodule,(III) rightdistalforearmwiththeproposedsensormoduleusingAg/AgClelectrodes, and(IV)leftdistalforearmwiththeproposedsensormoduleusingconductive fabricelectrodes. deviceisalsoinconspicuous,nonstigmatizing,andallowsfor discretemonitoringofEDA.Furthermore,theelectronicmod- ulecanbeeasilydetachedwhentheuserdesirestowashthe wristband. Todate,thereisnogenerallyacceptedstandardizationwith respecttoelectrodermalrecordingsites[3].Theelectrodesare commonlyplacedonthepalmarsurfaceofthehand,themost popularsitesbeingthemedialanddistalphalangesoftheÞn- gers,andthethenarandhypothenareminences(seeFig.4). However,sincebothhandsareoftenneededformanipulation, placementofelectrodesonthesesitesishighlysusceptibleto motionartifactsandinterfereswithdailyactivities.Thus,we decidedtousetheventralsideofthedistalforearmsasour recordingsites.WechosetouseAg/AgCldiscelectrodeswith contactareasof1.0cm 2 (ThoughtTechnology,Ltd.)forour recordings,asrecommendedintheliterature[33].Theseelec- trodesaredisposableandcanbesnappedontoorremovedfrom thewristbandwithease[seeFig.3(b)]. Overall,thecompletewearableEDAsensoriscompact (70mm × 70mm × 20mm),lightweight(40.3g),andthe componentsusedcanbepurchasedofftheshelfforapproxi- mately$150.Incontrast,acommercialsystemsuchastheFlex- compInÞniti(ThoughtTechnologies,Ltd.)measures130mm × 95mm × 37mm,weighs200g,andcosts$6000forthedata acquisitionunit,andanadditional$275foranEDAsensor. III.E XPERIMENTAL M ETHODS A.Participants Datawerecollectedfrom26participantsbetweentheages of18Ð56duringthreeseparateexperiments(physicaltask,cog- nitivetask,andemotionaltask)eachconsistingofabaseline, task,andrecoveryperiod.Someparticipantsunderwenttwo consecutiveexperiments(orderofexperimentswasnotÞxed), sotherewasoverlapbetweentherecoveryperiodoftheÞrst experimentandbaselineofthesecondexperiment.Thisstudy 1248 IEEETRANSACTIONSONBIOMEDICALENGINEERING,VOL.57,NO.5,MAY2010 Fig.7.AccuracyoftheproposedsensorinÞxed-resistancemeasurements. Errorbarsrepresent1S.D.Inset:Computedsensitivityoftheproposedsensor showingincreasingquantizationerrorathigherEDAvalues. 0.99)duringthebaseline,task,andrecoveryperiodforallthree experiments. Totesttheaccuracyoftheproposedsystem,wemeasuredase- riesof1%Þxedresistorsrepresentativeoftypicalskinresistance values(0.1Ð4.0M
)withtheproposeddeviceandcompared ourreadingstothatofadigitalmultimeter(0.5%accuracy). Theresultingerrorplotandcomputedsensitivityofthepro- posedsystemispresentedinFig.7.Themeasurementerrorof ourdevicewashigheratlowerresistancevalues,whichislikely duetotheincreasingquantizationerrorathigherconductance values(seeFig.7inset).Withinthetestedrangeofresistances, themeanmeasurementerrorwasfoundtobesmall(0.68 ± 0.64%).Acrossarangeoftypicalskinconductancevalues,the meansensitivityoftheproposeddevice(0.01 ± 0.01 µ S)was comparabletothatoftheFlexcomp(0.01 µ S). C.DistalForearmisaViableEDARecordingSite ToexaminethedegreetowhichdistalforearmEDAactiv- ityparallelsÞngerEDAactivity(atraditionalrecordingsite), wecorrelatedtheEDAsignalsfromthedistalforearmswith EDAsignalsfromipsilateralÞngers(IandIII,andIIandIVin Fig.4).Fig.8presentsthehistogramsofresultingcorrelation coefÞcientsforthetwodifferentelectrodetypesforallthree experiments( p 0.0001forallobservations).Ingeneral,the distributionswerenegativelyskewedwithlongtailstotheleft. BasedonthemedianofthecorrelationcoefÞcients  r ,correlation betweentheÞngersanddistalforearmwasverystrong(0.76
 r
0.96)duringthebaselineandrecoveryperiodsforallbutthe physicalandemotionalrecoveryperiodwiththeAg/AgClelec- trodes.TherewasalsoastrongcorrelationbetweentheÞngers anddistalforearmduringthephysical(  r = 0.78)andemotional (  r = 0.72)tasks.Duringthecognitivetask,thecorrelationswere lower,butremainedmoderatelystrong(  r = 0.57). ByusingAg/AgClelectrodesontherightdistalforearmand conductivefabricelectrodesontheleftdistalforearm,wealso comparedtheperformanceofthedifferentelectrodematerials. AlthoughthecorrelationbetweenÞngersanddistalforearmwas alsoverystrong(0.74
 r
0.88)duringbaselineandrecovery periodswiththeconductivefabricelectrodes,thecorrelation wasweakduringphysical(  r = 0.35)andcognitivetasks(  r = 0.45),andmoderatelystrongduringtheemotionaltask(  r = 0.57).Inaddition,wehereindeÞneaperformanceindex
tobe theproportionofobservationswithmoderatelystrong(0.5
r 0.8)toverystrong( r  0.8)correlationcoefÞcients.Ag/AgCl electrodesyieldedhighperformanceforallthreetasks(0.69


0.77),butforconductivefabricelectrodes,theperformance indexwaslower(0.46


0.54).Fromthisperspective, theAg/AgClelectrodesonceagainperformedbetterthancon- ductivefabricelectrodes,particularlyduringthestressortask periods.EvidencefromtheanalysisofcorrelationofEDAbe- tweenÞngersanddistalforearmssuggestthattherearelarge interindividualdifferenceswithasmallproportion(19%)of negativecorrelations. D.Long-TermInSituEDARecordingsRevealPatternsinAu- tonomicArousal Theparticipantwhoworetheproposedsensorforaweek reportednosideeffectsordiscomfort.FromFig.9,weseethat long-term insitu recordingscontainrichinformationaboutdaily patternsofskinconductancemodulation.Forexample,thereisa consistentpeakbetweenthehoursofmidnightand3am,corre- spondingtosleepactivity.NoSCLdriftwasobservableandthe recordingswererelativelyartifactfree,demonstratingtheabil- ityoftheproposedsensortocontinuouslymeasureEDAduring dailyactivities.Thebiggestadvantageoftheproposedwrist- wornAg/AgClsensoristheunprecedentedabilitytoperform comfortableandlong-termEDAmeasurements insitu ,captur- ingdifferencesinactivitiessuchassleep,studying,laboratory work,classwork,etc.,withouthavingtorecalibratebaselines andworryaboutwhetherlevelsarecomparable. V.D ISCUSSION Asanindexofsympatheticnervoussystemactivity,EDA offersimportantinsightintoabroadspectrumofpsychological andneurologicaldisorders.Theproposedsensorwristbandpro- videsapracticalandcomfortablenewsolutionforwidespread EDAassessment.UsingAg/AgClelectrodes,wefoundstrong correlationbetweentheproposedsensorwristbandandanFDA- approvedEDAmeasurementsystemacrossalltests.Thiscapa- bilityforlong-termEDAmeasurementsbothwithinandoutside ofalaboratoryorclinicalsettingcreatesexcitingopportunities forinvestigationsthatwouldotherwisehavebeendifÞcultto implement. Itisnotatrivialtasktodesignexperimentsforperformance comparisonbetweendifferentEDAsensors.Measurementsof ÞxedresistorsaloneareinsufÞcientanddonotprovideeval- uationofthedynamicpropertiesoftheEDAsensors.There- fore,thereisaneedtoperformtasksthatarewellknownto elicitchangesinEDAwhileperformingsimultaneousmeasure- mentswiththeproposedsystemandanestablishedsystem. However,bothpairsofelectrodescannotbeplacedonthesame recordingsiteduetocrosstalkinterference.Duetotheeffectof POH etal. :WEARABLESENSORFORUNOBTRUSIVE,LONG-TERMASSESSMENTOFELECTRODERMALACTIVITY1251 Fig.10.(a)CircuitforEDAmeasurements.ComputedproÞlesforthe (b)voltageand(c)currentßowacrossskinforarangeoftypicalskincon- ductancevalues. A PPENDIX ForexosomaticmeasurementsofEDAbydc,theconstant currentorconstantvoltagemethodiscommonlyused.However, thesetraditionalmethodsfacedifÞcultyinamplifyingtheEDA signalsinceinter-andintraindividualvariationsintonicSCLs resultinalargedynamicrange[3].Ontheotherhand,thephasic SCRsarerelativelysmall.InordertoachieveasufÞciently highresolutionoftheSCRsoveradynamicrangeofSCLs withouthavingtouncouplethetwocomponents,weemployed anautomaticbiascontrolmethodusingtwooperationampliÞers [seeFig.10(a)].TheÞrststagecomprisesanactivelow-pass Þlter(cutofffrequency f c = 1.6Hz)withvariablegain.To increasethedynamicrangeofmeasurements,thebias V b of theÞrstoperationalampliÞerisdeterminedbythefeedback fromtheoutputofthesecondstageintegrator V o (timeconstant  = 10ms).Theappliedvoltageacrosstheskindecreases inanonlinearfashionwithincreasingskinconductance[see Fig.10(b)].Althoughthecurrentßowthroughtheskinincreases nonlinearlywithskinconductance[seeFig.10(c)],thecurrent densityiswellbelowtherecommendedlimitof10 µ A/cm 2 [41]; thus,thereisminimalriskofdamagingsweatglands.Withina rangeofskinconductancebetween0.1and15 µ S,theaverage voltageappliedis0.47Vandtheaveragecurrentßowis2.37 µ A. Overall,theskinresistance R skin canbecalculatedasfollows: R skin =
V cc Š V b V b Š V o  R 1 . (1) EDAmeasuredasskinconductancecanbeobtainedsimply bytakingtheinverseof(1). A CKNOWLEDGMENT TheauthorswouldliketothankM.Feldmeier,K.Dobson, M.S.Goodwin,R.Fletcher,O.Wilder-Smith,Y.Kuboyama, E.Hedman,andS.Goyalfortheirvaluablefeedback,K.H.Kim foractivelyrecruitingparticipantsforthisstudy,andThought TechnologiesforgenerouslydonatingtheFlexcompInÞniti System. R EFERENCES [1]K.Sato,W.H.Kang,K.Saga,andK.T.Sato,ÒBiologyofsweatglands andtheirdisorders.I.Normalsweatglandfunction,Ó J.Amer.Acad. Dermatol. ,vol.20,pp.537Ð563,Apr.1989. [2]H.D.Critchley,ÒElectrodermalresponses:whathappensinthebrain,Ó Neuroscientist ,vol.8,pp.132Ð142,Apr.2002. [3]W.Boucsein, ElectrodermalActivity .NewYork:Plenum,1992. [4]T.Yokota,A.Sato,andB.Fujimori,ÒInhibitionofsympatheticactivity bystimulationoflimbicsystem,Ó Jpn.J.Physiol. ,vol.13,pp.138Ð144, Apr.15,1963. [5]T.YokotaandB.Fujimori,ÒEffectsofbrain-stemstimulationuponhip- pocampalelectricalactivity,somatomotorreßexesandautonomicfunc- tions,Ó Electroencephalogr.Clin.Neurophysiol. ,vol.16,pp.375Ð382, Apr.1964. [6]J.W.Papez,ÒAproposedmechanismofemotion,Ó J.Neuropsychiatry Clin.Neurosci. ,vol.7,pp.103Ð112,1937. [7]R.Edelberg,ÒElectrodermalrecoveryrate,goal-orientation,andaver- sion,Ó Psychophysiology ,vol.9,pp.512Ð520,Sep.1972. [8]S.M.VaezMousavi,R.J.Barry,J.A.Rushby,andA.R.Clarke,ÒArousal andactivationeffectsonphysiologicalandbehavioralrespondingduring acontinuousperformancetask,Ó ActaNeurobiol.Exp.(Wars) ,vol.67, pp.461Ð470,2007. [9]R.Vigouroux,ÒSurler ö oledelar « esistance « electriquedestissuesdans lÕ « electro-diagnostic,Ó Comp.Rend.Soc.Biol. ,vol.31,pp.336Ð339,1879. [10]F « er « e,ÒNotesurlesmodiÞcationsdelar « esistance « electriquesous lÕinßuencedesexcitationssensoriellesetdes « emotions,Ó Comp.Rend. Soc.Biol. ,vol.5,pp.217Ð219,1888. [11]J.Tarchanoff,ÒD « echarges « electriquesdanslapeaudelÕhommesous lÕinßuencedelÕexcitationdesorganesdessensetdediff « erentesformes dÕactivit « epsychique,Ó Comp.Rend.Soc.Biol. ,vol.41,pp.447Ð451,1889. [12]P.Bonato,ÒWearablesensors/systemsandtheirimpactonbiomedical engineering,Ó IEEEEng.Med.Biol.Mag. ,vol.22,no.3,pp.18Ð20, May/Jun.2003. [13]C.NiScanaill,S.Carew,P.Barralon,N.Noury,D.Lyons,andG.M. Lyons,ÒAreviewofapproachestomobilitytelemonitoringoftheelderly intheirlivingenvironment,Ó Ann.Biomed.Eng. ,vol.34,pp.547Ð563, Apr.2006. [14]P.S.Williamson,D.C.Fowles,andM.Weinberger,ÒElectrodermalpoten- tialandconductancemeasurementsclinicallydiscriminatebetweencystic Þbrosisandcontrolpatients,Ó Pediatr.Res. ,vol.19,pp.810Ð814,Aug. 1985. [15]A.M.MirkinandA.Coppen,ÒElectrodermalactivityindepression:Clin- icalandbiochemicalcorrelates,Ó Brit.J.Psychiatry ,vol.137,pp.93Ð97, Jul.1980. [16]A.M.Schell,M.E.Dawson,A.Rissling,J.Ventura,K.L.Subotnik, M.J.Gitlin,andK.H.Nuechterlein,ÒElectrodermalpredictorsoffunc- tionaloutcomeandnegativesymptomsinschizophrenia,Ó Psychophysi- ology ,vol.42,pp.483Ð492,Jul.2005. [17]M.P.Tarvainen,A.S.Koistinen,M.Valkonen-Korhonen,J.Partanen, andP.A.Karjalainen,ÒAnalysisofgalvanicskinresponseswithprinci- palcomponentsandclusteringtechniques,Ó IEEETrans.Biomed.Eng. , vol.48,no.10,pp.1071Ð1079,Oct.2001. [18]W.Hirstein,P.Iversen,andV.S.Ramachandran,ÒAutonomicresponses ofautisticchildrentopeopleandobjects,Ó Proc.Biol.Sci. ,vol.268, pp.1883Ð1888,Sep.22,2001. [19]N.Dolu,C.Ozesmi,N.Comu,C.Suer,andA.Golgeli,ÒEffectofhy- perglycemiaonelectrodermalactivityindiabeticrats,Ó Int.J.Neurosci. , vol.116,pp.715Ð729,Jun.2006. [20]P.DullerandW.D.Gentry,ÒUseofbiofeedbackintreatingchronic hyperhidrosis:Apreliminaryreport,Ó Brit.J.Dermatol. ,vol.103,pp.143Ð 146,Aug.1980. [21]Y.Nagai,L.H.Goldstein,P.B.Fenwick,andM.R.Trimble,ÒClinicalef- Þcacyofgalvanicskinresponsebiofeedbacktraininginreducingseizures inadultepilepsy:Apreliminaryrandomizedcontrolledstudy,Ó Epilepsy Behav. ,vol.5,pp.216Ð223,Apr.2004. [22]N.Pop-Jordanova,T.Zorcec,andA.Demerdzieva,ÒElectrodermal biofeedbackintreatingpsychogenicnonepilepticseizures,Ó Prilozi , vol.26,pp.43Ð51,Dec.2005. [23]Flexcomp,ÒFlexcompInÞniti,ÓNewYork,USA:ThoughtTechnologies Ltd.,2009. [24]SenseWear,ÒSenseWear,ÓPittsburgh,PA:BodyMedia,Inc.,2009. [25]QPET,ÒQPET,ÓAlmelo,Netherlands:BrainquiryB.V.,2009. [26]Vitaport,ÒVitaport,ÓKerkrade,Netherlands:TEMECInstrumentsB.V., 2009. IEEETRANSACTIONSONBIOMEDICALENGINEERING,VOL.57,NO.5,MAY2010 1243 AWearableSensorforUnobtrusive,Long-Term AssessmentofElectrodermalActivity Ming-ZherPoh ,StudentMember,IEEE ,NicholasC.Swenson,andRosalindW.Picard* ,Fellow,IEEE Abstract —Electrodermalactivity(EDA)isasensitiveindexof sympatheticnervoussystemactivity.Duetothelackofsensors thatcanbeworncomfortablyduringnormaldailyactivityand overextensiveperiodsoftime,researchinthisareaislimitedto laboratorysettingsorarticialclinicalenvironments.Wedevel- opedanovel,unobtrusive,nonstigmatizing,wrist-wornintegrated sensor,andpresent,fortheveryrsttime,ademonstrationoflong- term,continuousassessmentofEDAoutsideofalaboratorysetting. WeevaluatedtheperformanceofourdeviceagainstaFoodand DrugAdministration(FDA)approvedsystemforthemeasurement ofEDAduringphysical,cognitive,aswellasemotionalstressorsat bothpalmaranddistalforearmsites,andfoundhighcorrelations acrossallthetests.Wealsoevaluatedthechoiceofelectrodemate- rialbycomparingconductivefabricwithAg/AgClelectrodesand discussthelimitationsfound.Animportantresultpresentedinthis paperisevidencethatthedistalforearmisaviablealternativetothe traditionalpalmarsitesforEDAmeasurements.Ourdeviceoffers theunprecedentedabilitytoperformcomfortable,long-term,and insitu assessmentofEDA.Thispaperopensupopportunitiesfor futureinvestigationsthatwerepreviouslynotfeasible,andcould havefar-reachingimplicationsfordiagnosisandunderstandingof psychologicalorneurologicalconditions. IndexTerms —Electrodermalactivity(EDA),forearm,galvanic skinresponse,skinconductance,sweat,sympatheticnervoussys- tem,wearablesensors,wrist. I.I NTRODUCTION I NGENERAL,regulationofphysiologicalstatesofarousal isachievedbyabalanceofactivitywithinsympatheticand parasympatheticsubdivisionsoftheautonomicnervoussystem (ANS).Whiletheparasympatheticnervoussystempromotes restorationandconservationofbodilyenergy,thesympathetic nervoussystemstimulatesincreasedmetabolicoutputtodeal withexternalchallenges.Assuch,increasedsympatheticactiv- ity(sympatheticarousal)elevatesheartrate,bloodpressure,and sweating,aswellasredirectsbloodfromtheintestinalreservoir towardskeletalmuscles,lungs,heart,andbraininpreparation formotoraction.SympatheticpostganglionicÞbersconsisting ManuscriptreceivedJuly12,2009;revisedOctober18,2009.Firstpublished February17,2010;currentversionpublishedApril21,2010.Thisworkwas supportedbytheNancyLurieMarksFamilyFoundation. Asteriskindicates correspondingauthor . M.-Z.PohiswiththeHarvard-MITDivisionofHealthSciencesandTechnol- ogy(HST),Cambridge,MA02139USA,andalsowiththeMediaLaboratory, MassachusettsInstituteofTechnology,Cambridge,MA02139USA(e-mail: zher@mit.edu). N.C.SwensoniswiththeMediaLaboratory,MassachusettsInstituteof Technology,Cambridge,MA02139USA(e-mail:nswenson@mit.edu). *R.W.PicardiswiththeMediaLaboratory,MassachusettsInstituteof Technology,Cambridge,MA02139USA(e-mail:picard@media.mit.edu). ColorversionsofoneormoreoftheÞguresinthispaperareavailableonline athttp://ieeexplore.ieee.org. DigitalObjectIdentiÞer10.1109/TBME.2009.2038487 ofnonmyelinatedclassCnerveÞberssurroundeccrinesweat glandsandtheiractivitymodulatessweatsecretion[1].Since sweatisaweakelectrolyteandgoodconductor,theÞlling ofsweatductsresultsinmanylow-resistanceparallelpath- ways,therebyincreasingtheconductanceofanappliedcurrent. Changesinskinconductanceatthesurface,referredtoaselec- trodermalactivity(EDA),reßectactivitywithinthesympathetic axisoftheANSandprovideasensitiveandconvenientmea- sureofassessingalterationsinsympatheticarousalassociated withemotion,cognition,andattention[2].Stressisgenerally deÞnedasadisruptionoftheautonomicbalanceinvolvinga stateofhighsympatheticactivation.SinceEDAissolelyde- terminedbytheactivityofthesympatheticbranchoftheANS, whichispredominantinstressstates,tonicEDAparameters mayberegardedassuitablemeasuresofANSactivityinduced bystress[3].Thehypothalamus,whichisresponsibleforANS activity,playsamajorroleinelicitingipsilateralEDAunderthe directinßuenceoflimbicstructures[3].Theseinßuencesappear tostemfromantagonisticactionsoftheamygdala(excitatory) andhippocampus(inhibitory)[4],[5].Autonomicresponsesin theskinsuchassweating,piloerection,andvasomotorchanges canthusbeelicitedbyvariousemotionalstatesviathePapez circuitinthelimbicsystem[6].Inaddition,itiswidelyrecog- nizedthatattention-grabbingstimuliandattention-demanding tasksalsoevokeincreasedEDAresponses[7],[8]. Despiteimprovementsinmeasuringequipmentsincethe discoveryofelectrodermalphenomenamorethan100years ago[9]Ð[11],muchoftheresearchinthisareaislimitedto observationalmeasurementsperformedovershortperiodsof timeinlaboratorysettingsorartiÞcialclinicalenvironments. Theneedformonitoringpatientsoverextensiveperiodsoftime hasstimulatedinterestinwearabletechnologiesÑunobtrusive devicesthatcanbewornduringnormaldailyactivitytogather physiologicaldataoverperiodsofseveralweeksormonths[12]. Long-termmonitoringofEDAwillallowtheobservationof patternsofsympatheticarousalandregulationatasigniÞcantly longertimescale(daystomonths)comparedtoexistingstudies (minutestohours)andcouldpotentiallyrevealpreviouslyun- observabletrends.Inaddition,long-termmeasurementstaken inapersonÕsnaturalhomeenvironmentalsoprovideaclearer pictureofthepersonÕsphysiologicalstatethanashortperiodof assessmentinanunnaturalclinicalsetting[13].Clinically,wear- ableEDAsensorscanbeusedinpsychopathology,dermatology, andneurologyfordiagnosticpurposesandtherapyevaluation. PotentialclinicalapplicationsincludescreeningforcysticÞbro- sis[14],classiÞcationofdepressiveillnesses[15],predictionof functionaloutcomeinschizophrenia[16],discriminationbe- tweenhealthyandpsychoticpatients[17],characterizationof 0018-9294/$26.00©2010IEEE POH etal. :WEARABLESENSORFORUNOBTRUSIVE,LONG-TERMASSESSMENTOFELECTRODERMALACTIVITY1247 Fig.5.EDAwaveformsduring(a)physicalactivity,(b)cognitivestressors, and(c)emotionalstressors(ahorrormovieclip).Measurementswererecorded from(I)rightÞngerswiththeFlexcompsystem,(II)leftÞngerswiththepro- posedsensormodule,(III)rightdistalforearmwiththeproposedsensormodule usingAg/AgClelectrodes,and(IV)leftdistalforearmwiththeproposedsensor moduleusingconductivefabricelectrodes.(a)Physicalstressors(subject6). (b)Cognitivestressors(subject14).(c)Emotionalstressors(subject9). IV.R ESULTS A.EDAIncreasesDuringStressorTasks AllparticipantsreportedthatthewearableEDAsensors (wristbands)feltcomfortablethroughoutthedurationofthe study.ExamplesoftypicalEDAsignalsmeasuredduringthe threedifferentexperimentsarepresentedinFig.5.Inallthree experiments,skinconductancegraduallydecreasedtoaplateau duringtheinitialrelaxationperiodtoestablishabaseline.During thephysicaltask,skinconductanceincreasedastheparticipant wascycling[seeFig.5(a)].Inthecourseofthecognitivetask experiment[seeFig.5(b)],thestartofthementalarithmetictest (MAT)wasfollowedbyasteepriseinskinconductancelevel (SCL)thatremainedelevatedthroughoutthetest.Skinconduc- tanceincreasedtoahigherlevelduringtheStroopwordÐcolor matchingtaskin73.3%ofparticipants.Multipleskinconduc- tanceresponses(SCRs)werealsoevidentduringthetasks.From Fig.5(c),weseethatanSCRwasgeneratedwhentheinvestiga- torinformedtheparticipantthatahorrormoviewouldbeplayed. Atthestartofthemovie,theSCLsurgedupwardandremained highwithmultipleSCRsthroughoutthemovie.Whenthetasks endedandtheparticipantwasaskedtorelax,skinconductance decreasedsteadilytonearbaselineinallthreeexperiments. Fig.6.DistributionsofcorrelationcoefÞcientsbetweenEDAmeasurements frombilateralÞngers[Fig.4(IandII)]under(a)physical( n = 13),(b)cognitive ( n = 15),and(c)emotional( n = 13)stressors.Eachexperimentwasseparated intobaseline,task,andrecoveryconditionsforcorrelation.Themedianvaluesof thecorrelationcoefÞcients  r arealsopresentedforeachcondition.(a)Physical stressors.(b)Cognitivestressors.(c)Emotionalstressors. FromFig.5,wealsoseethatrecordingsfromoursensor modulesontheÞngers(II)anddistalforearmwithAg/AgCl electrodes(III)areinverycloseagreementwithrecordingsfrom theFlexcompontheÞngers(I).Overall,recordingsonthewrist withconductivefabricelectrodes(IV)producedthelowestskin conductancemeasurementsin86.7%ofparticipantsduringthe cognitivetaskand84.6%ofparticipantsduringtheemotional task.However,thefabricelectrodesproducedhighestreadings towardtheendofthephysicaltaskandduringthesubsequent recoveryperiodin46.2%ofparticipants. B.RecordingsofProposedSystemareHighlyAccurateand StronglyCorrelatedWithFDASystem Toevaluatetheperformancecharacteristicsofoursensor module,wecorrelatedthemeasuredEDAsignalsbythepro- poseddevicewiththeFlexcompfromtheleftandrightÞngers, respectively,(IandIIinFig.4)duringthebaseline,task,and recoveryconditionsforallthreeexperiments.Theresultinghis- togramsofthecorrelationcoefÞcients r aredisplayedinFig.6 ( p 0.0001forallobservations).Sincesomeofthedistributions wereslightlynegativelyskewedandthemeanisnotrobustto outlyingobservations,wechosetousethemedian  r asthemea- sureofcentraltendency.Overall,recordingsfromtheproposed systemandtheFlexcompwerestronglycorrelated(0.93
 r