BiomedicalSignalProcessingandControl31(2017)127–138
ContentslistsavailableatScienceDirect
Biomedical
Signal
Processing
and
Control
j ou rn a l h o m e p a g e :w w w . e l s e v i e r . c o m / l o c a t e / b s p c
Characterization
of
cochlear
implant
artifacts
in
electrically
evoked
auditory
steady-state
responses
Hanne
Deprez
a,b,∗,
Robin
Gransier
a,
Michael
Hofmann
a,
Astrid
van
Wieringen
a,
Jan
Wouters
a,
Marc
Moonen
baKULeuven,ExperimentalORL,Dept.Neurosciences,Herestraat49,3000Leuven,Belgium
bKULeuven,STADIUS,Dept.ElectricalEngineering(ESAT),KasteelparkArenberg10,3000Leuven,Belgium
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received18March2016
Receivedinrevisedform23June2016 Accepted26July2016
Keywords:
Cochlearimplant(CI) CIstimulationartifacts
Electricallyevokedauditorysteady-state responses(EASSR)
Linearinterpolation Monopolarmodestimulation
a
b
s
t
r
a
c
t
Objective:Electricallyevokedauditorysteady-stateresponses(EASSRs)areneuralpotentialsmeasured
intheelectroencephalogram(EEG)inresponsetoperiodicpulsetrainspresented,forexample,through
acochlearimplant(CI).EASSRscouldpotentiallybeusedforobjectiveCIfitting.However,EEG
sig-nalsarecontaminatedwithelectricalCIartifacts.Inthispaper,wecharacterizedtheCIartifactsfor
monopolarmodestimulationandevaluatedatwhichpulserate,linearinterpolationoverthesignalpart
contaminatedwithCIartifactissuccessful.
Methods:CIartifactswerecharacterizedbymeansoftheiramplitudegrowthfunctionsandduration.
Results:CIartifactdurationswerebetween0.7and1.7ms,atcontralateralrecordingelectrodes.At
ipsi-lateralrecordingelectrodes,CIartifactdurationsarerangefrom0.7tolargerthan2ms.
Conclusion:Atcontralateralrecordingelectrodes,theartifactwasshorterthantheinterpulseinterval
acrosssubjectsfor500pps,whichwasnotalwaysthecasefor900pps.
Significance:CIartifact-freeEASSRsarecrucialforreliableCIfittingandneuroscienceresearch.TheCI
artifacthasbeencharacterizedandlinearinterpolationallowstoremoveitatcontralateralrecording
electrodesforstimulationat500pps.
©2016TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-ND
license(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
Acochlearimplant(CI)isanelectronicdevicethatcanrestore hearingin severelyhearingimpairedsubjects.ACIsystem con-sistsofthreemainparts:anexternalspeechprocessor,theimplant, andanelectrode arrayinsertedinthecochlea.Thespeech pro-cessorconvertstheincomingsoundtoanelectricalstimulation pattern,whichistransmittedtotheimplantviaaradiofrequency (RF)link.Theelectrodesstimulatetheauditorynervewithbiphasic charge-balancedpulses[1].Twostimulationmodesareoftenused,
Abbreviations: AGF,amplitudegrowth function;C,maximum comfortable stimulationlevel;CI,cochlearimplant;d,interpolationduration;D,STIMartifact duration;EABR,electricallyevokedauditorybrainstemresponse;(E)ASSRs, (electri-callyevoked)auditorysteady-stateresponses;ECAP,electricallyevokedcompound actionpotential;I,interceptoftheCIartifactAGF;ICA,independentcomponent analysis;PCA,principalcomponentanalysis;POD,programmingdevice;RF,radio frequency;RFartifact,RFcommunicationlinkartifact;STIMartifact,electrical stim-ulationartifact;T,thresholdstimulationlevel;,slopeoftheCIartifactAGF.
∗ Correspondingauthorat:KULeuven,STADIUS,Dept.ofElectricalEngineering (ESAT),KasteelparkArenberg10bus2446,3001Leuven,Belgium.
E-mailaddress:hanne.deprez@esat.kuleuven.be(H.Deprez).
dependingonthereturnelectrode:bipolarmodeforstimulation betweenintra-cochlearelectrodesandmonopolarmodefor stim-ulationbetweenintra-andextra-cochlearelectrode(s).Inclinical settings, pulses are oftendelivered at highrates in monopolar mode,whichrequireslessbatterypowerthanstimulationin bipo-larmode.Furthermore,thresholdlevelsvarylessoverstimulation electrodes with stimulation in monopolar compared to bipolar mode,resultingineasierCIfitting.
Sinceearlyimplantationisprovencrucialforspeechand lan-guagedevelopment (e.g.[2]), anincreasingnumber of severely hearingimpairedinfantsreceiveaCIwithinthefirstyearoflife. PriortoCIactivation,thethreshold(T)andmaximumcomfortable (C)stimulationlevelsaredeterminedbasedonbehavioral(verbal) feedback.Thisisparticularlychallengingininfantsand subjects whocannotgivereliablebehavioralfeedback.Insuchcases, objec-tiveCIfittingbasedonelectrophysiologicalmeasurementscould beused.
Objective CI fitting based on electrophysiological measure-ments is currently under investigation. Transient responses to low-rate stimuli measured at the electrode-nerve interface (ECAPs)andatthebrainstemlevel(EABRs)havebeeninvestigated as objective measures for threshold estimation. However, the http://dx.doi.org/10.1016/j.bspc.2016.07.013
Fig.1.ExampleofaCIartifactforS8,withaCIattherightside,measuredwith37HzAM900ppspulsetrainsatasubthresholdstimulationamplitude.Left:timeand frequencydomainsignalsatrecordingelectrodesTP8(ipsilateral)andTP7(contralateral),referencedtoCz.Right:spatialdistributionofspectralpoweratthemodulation
frequency,referencedtoCz.TheunitsofthetopographyplotaredBnV=20log10nV,where1Vcorrespondsto60dBnVand0.1Vcorrespondsto40dBnV.Noneural
responseisexpectedtobepresent,assubthresholdstimulationlevelswereused.
thresholdvaluesobtainedwiththesemethodsthatuselow-rate stimuliareonlymoderatelycorrelatedwithbehavioralthresholds tohigh-ratepulsetrains[3–6].
ObjectiveCIfittingbasedonelectricallyevokedauditory steady-stateresponses(EASSRs)isalsobeingresearched.EASSRsareneural steady-stateresponsestoelectricalstimuliwithaperiodicity,such asa modulated pulsetrain. Theyare theelectrical analogue of auditorysteady-stateresponses(ASSRs),whichareevoked acous-tically,andcanberecordedwithheadmountedscalpelectrodes. ASSRsaretheresultofneuralphase-lockingtoanauditory stim-ulusand theresponse isbelieved toresultfromdifferentbrain regions,dependingontherepetitionormodulationfrequencyof thestimulus(furthercalledresponsefrequency)[7,8].(E)ASSRscan bedetectedinthefrequencydomainattheresponsefrequencyby meansofastatisticaltest,e.g.anF-testoraHotellingT2test[9,8]. EASSRsarecorruptedbyelectricalstimulationartifacts,which canbecausedbyboththeelectricalstimulationpulsesandtheRF communicationlinkbetweentheexternalspeechprocessorand theimplant.The former canhave a periodiccomponent atthe responsefrequencywhichmaydistorttheneuralresponse[12]. Fig.1showstheEEGsignalrecordedontwochannelsintimeand frequencydomain,forsubthresholdstimulation.BothEEGsignals haveacomponentatthemodulationfrequency,whichiscausedby theelectricalstimulationsincenoneuralresponseisbelievedtobe present.Thespatialdistributionofthespectralcomponentatthe modulationfrequencyisshowninthetopographyplot,indicating thattheelectricalstimulationartifactispresentonallrecording electrodes.Theamountofdistortionishighlysubject-dependent, andisaffectedbythestimulationparametersandtherecording electrode positions. Stimulation in monopolar mode results in largerCIartifactsthaninbipolarmode[10,11].
ItwasrecentlydemonstratedthatEASSRsinresponseto high-ratestimuliresultinelectrophysiologicalthresholdsthatcorrelate wellwithbehavioral thresholdsforstimulationinbipolarmode [12].Thenextstepistoevaluatethresholdestimationbasedon EASSRsforclinicallyusedparameters,inparticularforstimulation inmonopolarmode.
StimulationartifactscontaminatingtheEEGareaproblemin variousdomainswhereelectricalormagneticstimulationisused, includingdeepbrainstimulation,transcranialmagneticand cur-rentstimulation,somatosensoryandcochlearimplantstimulation. Changestothemeasurementset-up,suchasmaximum separa-tionofstimulationandrecordingelectrodeleads,propergrounding ofamplifierandsubject,andcarefulskinpreparationcanhelpto reduceartifactamplitudes[10,13].However,noneofthese meas-urescancompletelypreventthepresenceofexcessivestimulation artifactsintheEEG.Optimalreference electrodeplacement has beeninvestigatedfortransientresponsestocochlearimplant stim-ulation[14],butoptimalselectionofreferenceelectrodehasnotyet beenassessedforartifactremovalinEASSRmeasurements. Stimu-lusdesigncanalsohelptoavoidstimulationartifacts:responsesto alternatingpolaritypulseshavebeenaveragedinordertoreduce thestimulationartifact[15,16],orshortstimulihave beenused suchthatthestimulationartifacthasdecayedbeforetheresponse occurs[16].Adjustmentstothestimuliarenotdesirableinourcase, becauseweaimtomeasureEASSRstoclinicallyusedstimuli. There-fore,stimulationisrestrictedtocathodic-first,biphasicpulses,with fixedpulsewidthandinterphasegap,presentedathighratesand inmonopolarmode.
ArtifacteliminationmethodsremoveEEGchannelsorepochs thatarecontaminatedwithartifact.Thisisdoneforexamplewith ocularartifactsintheEEG.However,allepochsareaffectedby stim-ulationartifactsinEASSRmeasurementsbecauseofthecontinuous stimulation.Furthermore,mostrecordingchannelsareaffectedby stimulationartifact. Therefore, artifact eliminationmethods are notappropriateforartifactremovalinEASSRmeasurements,since almostalldatawouldberejected.
Severalmethods havebeenproposedforstimulationartifact minimization.Singlechanneltechniquesincludefrequency[17], time–frequency[18–20],oradaptivefiltering[21–26].Template subtraction [27–30] has also been investigated. In the case of EASSR,frequency domain filteringis inappropriatebecausethe stimulationartifacthasacomponentattheresponsefrequency. For adaptive filtering and template subtraction, assumptions
H.Deprezetal./BiomedicalSignalProcessingandControl31(2017)127–138 129
Fig.2.SimulatedCIartifactspectrumforunmodulatedpulsetrainspresentedatarepetitionfrequencyof40pps(left)andforhigh-rate(900pps)40HzAMpulsetrains (right),inthecaseofsymmetric(top)andasymmetricCIartifacts(bottom).
concerningthestimulationartifactshapeorfilteringprocessneed tobemade.
Interpolation methods [31,32,10,12] have also been used. Fortime-restrictedstimulationartifacts,aninterpolationcanbe appliedbetweenapre-artifactandpost-artifactsample,effectively removingthestimulationartifact.Thismethodisonlysuccessfulif theinterpulseintervalislargerthanthestimulationartifact dura-tion,andithasbeenvalidatedforEASSRmeasurementsinbipolar stimulationmode.
Multichannel techniques such as beamforming [33], prin-cipal (PCA) [16] and independent component analysis (ICA) [14,16,34–41]wereinvestigatedinvariousdomains.CIstimulation artifactshavesuccessfullybeenremovedfromtheEEGfor tran-sientresponsesusingmultichannelmethods,butthesemethods havenotyetbeeninvestigatedforsteady-stateresponses. Clini-cally,multichannelEEGsystemsareexpensiveandrequiremore subjectpreparationtime.
TheaimofthisstudyistocharacterizetheCIartifactfor mod-ulatedhigh-ratepulsetrainsstimulatedinmonopolarmodeand investigatethefeasibilityofstimulationartifactremovalwithlinear interpolation.Modulatedpulsetrainsareamodelforthe electri-calpulsesequencesafterprocessingofspeechbytheCIprocessor. LinearinterpolationwaschosenastheCIartifactremovalmethod, becauseitsefficiencyhasbeendemonstratedforbipolar stimula-tion,anditcanbeappliedtosinglechanneldata.TheCIartifact characterizationwillhelptoexplorethefeasibilityofotherabove mentionedCIartifactremovalmethods.Theinfluenceofreference electrodepositionontheCIartifactcharacteristicsandthe operat-inglimitsoftheinterpolationmethodwillbeinvestigated.
2. Materialsandmethods
TheCIartifactconsistsoftwomainpartsfromtheRF commu-nicationlink(theRFartifact)andfromtheelectricalstimulation (theSTIMartifact).TheCIartifactistime-lockedtotheelectrical
stimulationpulseandcancontainafrequencycomponentatthe modulationfrequency[10,12],ascanbeseeninFig.1forrecorded dataandFig.2forsimulatedcases.Thismayresultindistorted EASSRpropertiessuchasamplitudeandphaseandfalsepositive EASSRdetections.
InCochlearNucleus®implants,thestimulationamplitudeofthe pulsesisnonlinearlyencodedintheRFtransmissionandis there-foreconstantforstimulationpulseswithdifferentamplitudes[28], whereastheSTIMartifactamplitudeisrelatedtothestimulation pulseamplitude.Forstimulationwithunmodulatedpulsetrains, boththeRFandtheSTIMartifactarepresentattheresponse fre-quency(namelytherepetitionfrequencyofthestimulationpulses). For stimulationwithhigh-rate modulatedpulsetrains,onlythe STIMartifactispresentattheresponsefrequency(namelythe mod-ulationfrequencyofthestimulationpulses).Furthermoreifthe STIMartifactissymmetric,noSTIMartifactwillbepresentatthe responsefrequency,ascanbeseeninFig.2.Inthefollowing,we onlyconsiderstimulationwithhigh-ratemodulatedpulsetrains. Inthis case,onlytheSTIMartifactcomponentsareproblematic forEASSRmeasurementsastheymayhaveacontributionatthe modulationfrequency.
ThescalingoftheCIartifactswithincreasingstimulation ampli-tudeisquantifiedbymeansoftheslopeoftheCIartifactamplitude growthfunction(AGF).Iftheslopeiszero,theCIartifactsdonot scalewithchangingstimulationamplitude,whichindicatesthat theywillnotbepresentatthemodulationfrequency.IfCI arti-factsarepresentatthemodulationfrequency,theycanpossiblybe removedwithalinearinterpolation.However,thisonlyworksif theCIartifactisshorterthantheinterpulseinterval.Therefore,the STIMartifactdurationisalsoquantified.
EASSRsweremeasuredin11subjectswithaCochlearNucleus® CI with stimulation below the subject’s behavioral threshold level.Detailsaboutsubjects,stimulation,andrecordingsetupare describedinSections2.1,2.2and2.3,respectively.CIartifactAGF interceptsandslopesandSTIMartifactdurationsweredetermined
Table1
ListofsubjectswithCochlearNucleus®implantdetails.S:subjectidentifier;Sex:M:
male,F:female;Age:ageinyears;Exp:CIexperienceinyears;Sideofimplantation: R:right,L:left;PR:pulseratetested.
S Sex Age Exp Implanttype Side PR
500pps 900pps S1 F 55 16 CI24R L x S2 M 64 11 CI24R L x x S3 M 19 17 CI24M L x S4 F 85 5.7 CI24R L x x S5 M 74 1.2 CI24RE R x S6 M 52 1.7 CI24RE R x S7 M 64 16 CI24R L x x S8 M 52 1.9 CI24RE R x x S9 F 44 0.5 CI422 R x x S10 F 77 1.9 CI24Re L x S11 F 63 2.5 CI24RE R x
forallsubjectsasdescribedinSection2.4.Allsignalprocessingand
statisticalanalysesweredoneinMATLABR2013a. 2.1. Subjects
Intotal,11adultsubjectsparticipatedintheexperiments.They allhad aCochlearNucleus® CI.DetailscanbefoundinTable1. Allsubjectstook partvoluntarily and signed an informed con-sentform.Theexperimentswereapprovedbythemedicalethics committeeoftheUniversityHospitalsLeuven(approvalnumber B32220072126).
2.2. Stimulationsetup
Anin-housedevelopedstimulationsoftwareplatform gener-ated the electrical stimulation pulse sequences with specified stimulationparameters,suchaspulserate,modulationfrequency, stimulationelectrode,etc.[10].Theelectricalpulsesequenceswere senttoaprogrammingdevice(POD)connectedtoaL34research speechprocessorprovidedbyCochlearLtd.,therebybypassingthe subject’sclinicalspeechprocessor.
CochlearNucleus® implantshave tworeturn electrodes out-sidethecochlea,i.e.thecasingandtheballelectrode.Allsubjects werestimulatedinmonopolarmodeMP1+2,i.e.betweenan intra-cochlear electrode and thetwo extracochlear return electrodes whichareelectricallycoupled[42].Anintracochlearelectrodein themiddleofthearraywasused:electrode11wasusedforall subjectsexceptS1,forwhomelectrode13wasused.Thestimuli consistedofamplitude-modulated(AM)pulsetrainswith modula-tionfrequenciesinthe40Hz-range,whichisoftenusedfortesting adultsbecauselargeresponsesareexpectedhere.Clinicallyused symmetricbiphasicpulseswithapulsewidthof25sandan inter-phasegapof8swereusedforstimulation.
Thresholdandcomfortlevelsweredeterminedforstimulation withunmodulated(Tu andCu)andAMpulsetrains(TmandCm). TheTlevelisthestimulationamplitude(inCochlearclinicalcurrent units(cu),aunitofelectricalcurrent)thatelicitsajustperceivable auditoryperception.TheClevelis thestimulationamplitudeat perceivedmaximumcomfortableloudness.For AMpulsetrains, thedeterminedTmandCmrefertothemaximumamplitudeofthe AMpulsetrainsthatresultinajustperceivableauditorysensation andamaximallycomfortablesound,respectively.
Twostimulationpulseratesweretested:500ppswhichisatthe lowerendofclinicallyusedstimulation,and900ppswhichisthe defaultpulserateusedinCochlearNucleus®implants.Thestimuli weremodulatedwithfrequenciesinthe40Hzrange.Subjectswere stimulatedatsubthresholdstimulationpulsetrainintensities,with modulationdepthequalto(Cm−Tu)/(Cm+Tu),during5min.
2.3. Recordingsetup
Tostudytheeffectofreferenceelectrodeposition,a64-channel active-electrodeBioSemiActiveTwoDCEEGrecordingsystemwas used.Thesystemhasa24bitresolutionoveradynamicrangeof 524mVPPandasamplingrateof8192Hzwasused.Therecording setuphasabuilt-inanalog5thordersinclow-passfilterwitha cutofffrequencyof1638Hz.Recordingelectrodeswereplacedon thesubject’sheadaccordingtothepositionsoftheinternational 10–20system[43].Atriggersignalwassenttotherecordingsystem forsynchronizationatthebeginning ofeachrecordingepochof 1.024s.AfterEEGsignalrecordingduring5min,thesignalswere rereferencedofflinetothreecommonlyusedreferenceschemes: averagereference,vertexreferenceCz,andforeheadreferenceFpz. The recordings were made in a soundproof and electrically shieldedroom.Subjectswereseatedina comfortablechairand wereaskedtomoveaslittleaspossible.Asilentbutsubtitledmovie oftheirchoicewasplayed,toguaranteethesameattentionalstate acrosssubjectsandmeasurements.
2.4. CIartifactcharacterization 2.4.1. CIartifactamplitudegrowth
TheCIartifactAGFA(As)showshowtheCIartifactamplitudeA changeswithincreasingstimulationpulseamplitudeAs.
TheCIartifactamplitudeApwasdeterminedforevery stimu-lationpulse.Letxp(t,c)betheEEGsignalfollowingpulsep(with stimulationamplitudeAs(p)A)attimetandchannelc.Inthe fol-lowing,wewillmakeabstractionofthechannelcasthemethod canbeappliedtoeverychannelseparately.Ap(inV)wasdefined asthesumofthepulse’smaximalandminimalamplitude. Ap=
maxt xp(t)+mint xp(t)
(1)For symmetrical artifacts, with equal negative and positive amplitudes,Apwillbezero.Forasymmetricalartifacts,Apwilldiffer fromzero.
For each stimulation pulse pstimulated at amplitude As(p), themaximalandminimalEEGamplitudesweredeterminedand summed,resultinginAp.Next,thesevalueswereaveragedforall pulsespresentedatthesamestimulationamplitude,suchthatone CIartifactamplitudeAisdeterminedforeach stimulationpulse amplitudeAs.Inafirstapproximation,theCIartifactAGFA(As)can bemodeledasalinearfunctionofAs:A(As)=mAs+Iwithintercept Iandslopem=◦,asshowninFig.3.Thebestlinearfitwas deter-minedforeverychannelwithaleastsquaresprocedure,resulting invaluesfortheinterceptIandslope.
TheinterceptIrepresentsasymmetricCIartifactcomponents thatareconstantacrossstimulationpulseintensities;these arti-factcomponentsaremainlycausedbytheRFartifact.Theslope representsasymmetricCIartifactcomponentsthatchangewith increasingstimulationpulseamplitude,namelytheSTIMartifact. IftheCIartifactissymmetric,bothandIwillbezero.Ifiszero andIisnon-zero,theCIartifactismainlycausedbyRF transmis-sion.IfbothandIarenon-zero,theCIartifactconsistsofRFand STIMartifact.OnlytheSTIMartifactcomponentsareproblematic forEASSRmeasurementsasthesearetheonlycomponentsthat haveacontributionatthemodulationfrequency.
Aslopeof1◦correspondstoanincreaseof0.017V/A.Inthis study,thestimulationamplituderangeaveragedoversubjectsis about100A.Therefore, the amplitudedifferencebetweenthe largestandsmallestpulseamplitudefor=1◦is1.7Vforan aver-agesubject.Thisisalargevalue,comparedtotheneuralresponse whichhasamplitudesbetween20and500nVforaveragesubjects inthe40Hzrange[12].
H.Deprezetal./BiomedicalSignalProcessingandControl31(2017)127–138 131
Fig.3. CIartifactAGFsforS1andS8,measuredwith37HzAM900ppspulsetrainsatasubthresholdstimulationamplitude,betweenanipsilateraloccipitalelectrode(O2)
andforeheadreferenceelectrode(Fpz).
CIartifactAGFswereconstructedforallsubjects.Examplesof suchAGFsareshowninFig.3.Pulseratesof500and900ppswere used,althoughnotallsubjectsweretestedwithstimulationatboth pulserates,asshowninTable1.TheDCbiaswasremovedfromthe recordedEEGsignalswithasecond-order2Hzhigh-passfilterand theEEGsignalswererereferencedtoeitheraveragereference,Cz, orFpz.ThevaluesoftheinterceptIandslopeoftheCIartifact AGFweredeterminedforeveryrecordingchannelandfordifferent recordingelectrodeconfigurations.
2.4.2. STIMartifactduration
Artifactswereremovedbylinearinterpolationbetweena pre-stimulusandpost-stimulussample.Thetimebetweenthepre-and post-stimulussampleis calledtheinterpolationdurationd.The maximumpossibleinterpolationdurationisdefinedasthe inter-pulseinterval,whichistheinverseofthepulserate,andequals 2msand1.1msforstimulationat500and900pps,respectively. Inthiscase,onesampleperpulseperiod,thepre-stimulus sam-ple,isretained.Alinear interpolationwasappliedbetweenthe pre-stimulussampleat−100sandpost-stimulussamples vary-ingbetween+500and+1900s,instepsof100sfor500pps.For 900pps,post-stimulussamplesvaryingbetween+500and+900s, wereused,instepsof100s.Thesamplingrateisnotanexact mul-tipleofthepulserate.Therefore,thestartofastimulationpulseis notexactlyalignedtoasample.Thestartandendsamplesofthe interpolationarecalculatedforeachpulseseparately,byrounding thestartandendtimeoftheinterpolationtothenearestsample. Lookingoverthewholerecording,theaveragetimebetweenthe startoftheinterpolationintervalandthestartofastimulationpulse isequaltothepre-stimulusinterpolationduration.Equivalently, theaveragetimebetweenthestartofastimulationpulseandthe endoftheinterpolationintervalisequaltothepost-stimulus inter-polationduration.Post-stimulussamplesbefore+500swerenot used,astheCIartifactpeaklastsforabout500s,ascanbeseenin Fig.4.
After linear interpolation, the signals were filtered with a second-order2Hzhigh-passfilter,rereferencedtoeitheraverage reference,CzorFpz,andsplitinto1.024sepochs.The300 result-ingepochs,correspondingtoa5minrecordinglength,werethen averagedtoreducethenoiseleveln.Then,theresultingspectral amplitudesAmatthemodulationfrequencyinfunctionofthe inter-polationdurationdweredetermined,asillustratedinFig.4.
WhentheinterpolationdurationisshorterthantheSTIM arti-fact,Am(d)maystillcontainsomeSTIMartifact.However,Am(d) decreaseswithincreasinginterpolationduration,asalargerpart oftheSTIMartifactisthencanceled.Whentheinterpolation dura-tionislongerthantheSTIMartifactduration,Am(d)stabilizesat theneuralresponseamplitude,namelytherealEASSRamplitude. Am(d)stabilizestothenoiselevel,inourcase,asnoneuralresponse isexpectedtobepresentforsubthresholdstimulation.
AnAm(d)AGFexampleisshowninFig.4.ThedifferencesinAm forincreasinginterpolationdurationdwerecomparedtothenoise levelafteraveragingn.TheSTIMartifactdurationDwasdefinedas theshortestinterpolationdurationforwhichthisdifferencedidnot exceedthesubjectdependentnoiseleveln,whichisapproximately 50nV:
D=d:[Am(d)−Am(d−1)]<n (2)
IfAm(d)didnotsaturate,meaningthatthedifferenceinAm(d)was notsmallerthanthenoiselevelforanyinterpolationdurationd,the STIMartifactdurationDwassetequaltothemaximalinterpolation duration.
2.4.3. Statisticalanalyses
TheinterceptIandslopeoftheCIartifactAGFandtheSTIM artifact duration Dwere determinedas described above for all recording electrodes and for three reference electrode configu-rations inall subjects.Leftandright recordingelectrodes were switchedforsubjectswithaCIattherighthandside,toputthe resultsinthesamefigureforsubjectswithaCIattheleftandright
0.5 0.7 0.9 1.1 1.31.5 1.7 1.9 0 0.05 0.1 0.15 0.2 0.25 Spectral amplitude at f m A m (uV) 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 0 0.05 0.1 0.15 0.2 0.25 −600 −500 −400 −300 −200 −100 0 Amplitude (uV) Ipsilateral −50 0 50 100 Contralateral
Post−stimulus interpolation duration d (ms)
Fig.4.CIartifactpulse(top)andAm(d)AGFwithincreasinginterpolationdurationdforsubjectS2(bottom).StimulationbelowTlevelat500pps,foripsi-andcontralateral
recordingelectrodes.TheSTIMartifactdurationsareindicatedindash-dottedlines.ReferenceelectrodeCz.
handside.Theresultingsignalswereaveragedacrossallsubjectsto obtaintheaverageCIartifactprofileshowninFig.5.Thismaygive ablurredview,asCIartifactsmaybelocalizedslightlydifferently inallsubjects.
Inthefollowing,onlyrecordingelectrodeslocatedinthe pos-teriorpartofthehead(Tx,C(P)x,P(O)x,Ox,Ix)wereconsidered.For eachsubject,themedianvalueof,IandDovertherecording elec-trodeswasdetermined.Astatisticalanalysisinvestigatingtheeffect ofreferenceelectrode,hemisphere,andpulserateon,IandDwas
carriedout.Alleffectsarereportedatasignificancelevelof5%. ThedatawerenotnormallydistributedaccordingtoaJarque–Bera test,andthereforeonlynonparametrictestswereused.AFriedman analysiswasusedtoinvestigatetheeffectofreferenceelectrodeon theCIartifactAGFslopeandSTIMartifactdurationforeachpulse rateandforeachhemisphere.Theeffectofhemispherewas inves-tigatedusingWilkinsonsignedranktests(averagingtheresultsfor referenceelectrodesCzandFpz).TheinfluenceofpulserateonCI artifactAGFslopeandSTIMartifactdurationwascheckedforeach
Fig.5. MeanslopeandinterceptIoftheCIartifactAGFandmeanSTIMartifactduration,averagedoverallsubjectswithrecordingswithstimulationat500pps.Average reference(leftcolumn)andreferenceelectrodeCz(rightcolumn).
H.Deprezetal./BiomedicalSignalProcessingandControl31(2017)127–138 133 −10 0 10 20 30 40 50 60 Ipsilateral
CI artifact AGF slope θ (degrees)
0 200 400 600 800 1000
CI artifact AGF intercept I (uV)
S2 S4 S6 S7 S8 S9 S10 S11 0 2 4 6 8 10 Contralateral S2 S4 S6 S7 S8 S9 S10 S11 0 200 400 600 800 1000 Subject
Fig.6.CIartifactAGFslopeandinterceptIforipsi-andcontralateralposteriorrecordingelectrodesforeachsubjectwithrecordingswithstimulationat500pps.Reference electrodeCz.Theboxplotshowsthemedianand25th(q1)and75thpercentiles(q3).Outliers(+)arealldatapointsthatfalloutsidetherange [q1±1.5(q3−q1)] .
hemisphere(averagingtheresultsforreferenceelectrodesCzand Fpz),usingWilkinsonranksumtests.
3. Results
3.1. CIartifactAGFslopeandintercept
CIartifactAGFslopesandinterceptsareshowninFigs.6and7. TheCIartifactissymmetricifbothandIarezeroandthisisonly thecaseforsubjectS1(Fig.7).
MostsubjectshadaCIartifactAGFsimilartothatofsubjectS8 inFig.3.TheCIartifactslopeisdifferentfromzero,whichmeans thattheSTIMartifactcontributestotheCIartifact.
3.2. STIMartifactduration
Fig.8 showsSTIM artifactdurationsDfor each subject sep-arately, on all ipsilateral and contralateral posterior recording electrodes respectively. For stimulation at 500pps, the median STIM artifact duration at the ipsilateral posterior electrodes is 1.6ms,althoughtheSTIMartifactdurationisclosetoorlongerthan 2msatsomeelectrodesinsomesubjects(Fig.8).Forstimulation at900pps,thedeterminedSTIMartifactdurationisequalto1.1ms inalmostallsubjectsatipsilateralrecordingelectrodes.Disthus largerthanthemaximumpossibleinterpolationduration.
Atcontralateralrecordingelectrodes,themedianSTIMartifact durationis1and0.9msat500and900pps,respectively.For stim-ulationat900pps,Discloseorequalto1.1msinsomesubjects (Fig.8).
3.3. Influenceofreferenceelectrodeandhemisphere
TheslopeandinterceptIoftheCIartifactAGFandtheSTIM artifact duration D are largest in theproximity of the implant (Fig.5).,IandDarelargerinthecontralateralhemispherefor averagereferencethanforreferenceelectrodeCz.
For stimulation at 500pps, a significant influence of ref-erence electrode on CI artifact AGF slopes was found in the ipsilateral(2(2)=7.8,p=0.021)andthecontralateralhemisphere (2(2)=6.8, p=0.034), see Fig. 9. In the ipsilateralhemisphere, largerCIartifactAGFslopeswerefoundfortheFpzreference elec-trodemontage. In thecontralateralhemisphere,more variation inCIartifactAGFslopeisobservedwhenreferenceelectrodeCz isusedcomparedtowhenreferenceelectrodeFpzischosen.For stimulationat900pps,asignificantinfluenceofreferenceelectrode onCIartifactAGFslopeswasfoundintheipsilateralhemisphere (2(2)=9,p=0.011)andinthecontralateralhemisphere(2(2)=7, p=0.030).
Forstimulationat500pps,asignificantinfluenceofreference electrodeonCIartifactAGFinterceptwasfoundinthe contralat-eral (2(2)=7, p=0.030), but not in the ipsilateral hemisphere (2(2)=1.0,p=0.607),seeFig.10.Inthecontralateralhemisphere, smallerCIartifactAGFinterceptswerefoundfortheFpzreference. For stimulationat 900pps, nosignificantinfluenceofreference electrodeonCIartifactAGFinterceptswasfoundintheipsilateral hemisphere(2(2)=1.75,p=0.417)orinthecontralateral hemi-sphere(2(2)=0.75,p=0.687).
Forstimulationat500pps,thereferenceelectrodewasfoundto haveasignificantinfluenceonSTIMartifactdurationonipsilateral (2(2)=6.1, p=0.048) and contralateral electrodes (2(2)=14.6, p<0.001).Inthecontralateralhemisphere,shorterSTIMartifact durationswerefoundfortheFpzreference.Intheipsilateral hemi-sphere,the STIMartifact duration is largerthan themaximum possibleinterpolationdurationforstimulationat900pps. There-fore the influence of reference electrode on the STIM artifact durationwasonlycheckedinthecontralateralhemisphere.The ref-erenceelectrodewasfoundtohaveasignificantinfluenceonSTIM artifact durations in the contralateral hemisphere (2(2)=13.0, p=0.002).ShorterSTIMartifactdurationswereagainfoundforthe Fpzreference.
For stimulation at 500pps there was a significant effect of hemisphere on CI artifact AGF slope, offset and STIM artifact duration (p=0.008, p=0.008, and p=0.008, respectively). Prior
−10 0 10 20 30 40 50 60 Ipsilateral
CI artifact AGF slope θ (degrees)
0 200 400 600 800 1000
CI artifact AGF intercept I (uV)
S1 S2 S3 S4 S5 S7 S8 S9 0 2 4 6 8 10 Contralateral S1 S2 S3 S4 S5 S7 S8 S9 0 200 400 600 800 1000 Subject
Fig.7.CIartifactAGFslopeandinterceptIforipsi-andcontralateralposteriorrecordingelectrodesforeachsubjectwithrecordingswithstimulationat900pps.Reference electrodeCz.
tothestatisticalanalysisresultsfor referenceelectrodesCz and Fpz were averaged. For these reference electrodes, the CI arti-fact AGF slope is smaller and STIM artifact duration is shorter in the contralateral hemisphere. For stimulation at 900pps, a significanteffectofhemispherewasfoundontheCIartifactAGF slope and intercept (p=0.016 and p=0.016, respectively),with smallerCIartifactAGFslopesandinterceptsinthecontralateral hemisphere. The effect of hemisphere on STIM artifact dura-tioncould not beinvestigated, as theSTIM artifact duration is
longerthanthemaximalinterpolationdurationintheipsilateral hemisphere.
3.4. Influenceofpulserate
NosignificantinfluenceofpulserateonCIartifactAGFslope wasfound,forneitherofthehemispheres(p=0.798andp=0.721 foripsi-andcontralateralhemisphere,respectively).Nosignificant influenceofpulserateonCIartifactAGFinterceptwasfound,for
0.6 0.8 1 1.2 1.4 1.6 1.8 2 500 pps Ipsilateral 900 pps S2 S4 S6 S7 S8 S9 S10 S11 0.6 0.8 1 1.2 1.4 1.6 1.8 2 Contralateral S1 S2 S3 S4 S5 S7 S8 S9 Subject
STIM artifact duration D (ms)
Fig.8.STIMartifactdurationforipsi-andcontralateralposteriorrecordingelectrodesforeachsubjectwithrecordingswithstimulationat500and900pps.Reference electrodeCz.Dottedlinesindicatetheminimumandmaximumpossibleinterpolationdurationat500and900pps.
H.Deprezetal./BiomedicalSignalProcessingandControl31(2017)127–138 135
Fig.9. CIartifactAGFslopeperpulserate(500and900pps),hemisphere(ipsi-andcontralateral)andreferenceelectrode(averagereference,CzandFpz).Thesymbols*,
**,and***indicatethatp-valuesaresmallerthan0.05,0.01,and0.001,respectively. neitherofthehemispheres(p=0.235andp=0.328foripsi-and contralateralhemisphere,respectively).Nosignificantinfluenceof pulserateonSTIMartifactdurationwasfoundinthe contralat-eralhemisphere(p=0.343).STIMartifactdurationsforbothpulse ratescouldnotbecomparedintheipsilateralhemisphere,asthe STIMartifactdurationexceededthemaximumpossible interpola-tiondurationat900pps.
4. Discussion
Inthisstudy,theCIartifactwascharacterizedbasedonthree properties,namely, theCI artifactAGF slope and intercept and theSTIM artifact duration. The CI artifact AGF slope and STIM
artifactdurationdescribehowtheCIartifactscaleswithstimulation amplitudeandhowlongittakestheSTIMartifacttohavedecayed completely,respectively.SignificantlylargerCIartifactAGFslopes andinterceptsandSTIMartifactdurationsarefoundatipsilateral recordingelectrodesthanatcontralateralones.Forelectrodes pos-itionedatthecontralateralside,thereferenceelectrodelocation canhaveaninfluenceontheCIartifactAGFslopeandintercept(for stimulationat500pps)andSTIMartifactduration.Nosignificant influenceofpulserateonanypropertyhasbeenfound.Basedon theSTIMartifactdurations(Fig.8),itshouldbepossibletoremove STIMartifactsatcontralateralelectrodeswithalinear interpola-tionforstimulationat500pps.Forstimulationat900pps,more advancedmethodsareneeded.
Fig.11.STIMartifactdurationperpulserate(500and900pps),hemisphere(ipsi-andcontralateral)andreferenceelectrode(averagereference,CzandFpz).Dashedand
dottedlinesindicatethemaximumpossibleinterpolationdurationat500and900pps,respectively.Thedash-dottedlineindicatestheminimuminterpolationdurationused fortheanalysis.
Itisnotrecommendedtouseaveragereferencesubtractionwith CIstimulation.Whenthesubtractedreferencecontainsmore arti-factthanthechannel fromwhichit issubtracted,theresulting signalinthatchannelcouldcontainmoreCIartifactafter refer-encesubtractionthanbefore.LargeCIartifactsignalspresentat someelectrodeswillbiasthemeansignaloverchannels,resulting inlargeCIartifactsatallchannelsafterreferencesubtraction.
ThereferenceelectrodehasnotonlyaninfluenceontheCI arti-factcharacteristics,but alsoonthedetectedEASSR.Thesource oftheEASSRsisorientedalongadipole.Inordertorecord reli-ableEASSRswithmaximalamplitudes,theanalysisandreference electrodesshouldbeplacedonoppositesidesalongand perpen-diculartotheaxisofthisdipole.ThelocationoftheEASSRsource inthebrainvarieswithvaryingmodulationfrequencies.EASSRs tomodulationfrequenciesinthe40Hzrange(20–60Hz)originate fromsub-corticalsources[7].Whetheritispossibletoadequately recordEASSRswithaspecificcombinationofanalysisandreference electrodesthus dependsontheselected modulationfrequency. EASSRswerealsorecordedatsuprathresholdstimulationlevelsfor thesamesubjects,andforthemodulationfrequencieswetested inthe40Hzrange,itwasstillpossibletorecordreliableEASSRs whenreferenceelectrodeFpzwasselected(datanotpresented). ReferenceelectrodeFpzisalsooftenusedinclinicalABRandASSR measurementsininfants[44].Itshouldbenotedthat referenc-ingthedatatoFpz canleadtoincreasednoise levels,resulting in reduced EASSR detectionsor requiring longer measurement times.
OnlyonesubjecthadsymmetricCIartifacts,thatdidnotscale withincreasingstimulationamplitude.CIartifactshada contribu-tionatthemodulationfrequencyforallothersubjects.
TheamplitudeAmatthemodulationfrequency,consistingof contributions from the STIM artifact and the neural response, reduceswithincreasinginterpolationduration.TheAmdifference forsubsequentinterpolationdurationswascomparedtothenoise levelafteraveraging.TheSTIMartifactdurationisthe interpola-tiondurationforwhichthisdifferencebecomessmallerthanthe noiselevel.BecausewelookatthesaturationofAm,andnotat itsabsolutelevel,this methodcanalsobeappliedtorecordings
withstimulationatsuprathresholdlevels.Furthermore,thetimeT overwhichtheEEGsignalswereaveragedplaysanimportantrole here,sincethenoiselevelisdependentonthis.Thenoiselevelis reducedwithafactor√2eachtimetheaveragingtimeisdoubled. TheSTIMartifactdurationthusdetermineswhethertheSTIM arti-factcanberemovedbylinearinterpolationforrecordingtimeT.For longerrecordingtimes,thecontributionoftheSTIMartifactmay notbebelowthenoiselevelandtheSTIMartifactispossiblynot completelyremovedbyapplyinglinearinterpolation.
Forstimulationat500pps, theSTIMartifactcanberemoved atcontralateralrecordingelectrodeswithalinearinterpolation.In thecontralateralhemisphere,thevariabilityof(forstimulation at 500pps) and Dwassmallerfor more frontalreference elec-trodes,seeFigs.9and11.Therefore,forrecordingelectrodesinthe contralateralhemisphere,thechancethattheSTIMartifact dura-tionislongerthanthemaximumpossibleinterpolationdurationis reducedbychoosingamorefrontalreferenceelectrode.
Linearinterpolationisnotsufficienttoexamineresponsesat ipsilateralrecordingelectrodesorforstimulationpulserateshigher than500pps.Otherstimulationartifactremovalmethodsshould therefore be examined. Further modeling of CI artifacts could allowtemplatesubtractionoradaptivefilterdesignforCIartifact removal.Multichannelmethodscouldpossiblybeused,withthe disadvantage–forCIfittingpurposes–thattheserequireamore expensivesetupandmoresubjectpreparationtime.
OnlysubjectswithCochlearNucleus®implantsparticipatedin thisstudy.However,thestimulationartifactscausedbyimplants fromothermanufacturersshouldbeexamined,usingthemethods presentedhere.Differencescanbeexpected,asother manufactur-ersusedifferentclinicalparameters.
Inthisstudy,a64-channelrecordingset-upwasused,which allowedtoinvestigatetheinfluenceofreferenceelectrode posi-tionontheCIartifactcharacteristics.Setupswithlesschannelscan usuallybeoperatedathighersampleratesandhavelowpass fil-terswithhighercut-offfrequencies,whichcouldresultinshorter STIMartifactdurations.Themethodpresentedinthisstudycanstill beusedtodeterminetheSTIMartifactdurationandtherequired interpolationduration.
H.Deprezetal./BiomedicalSignalProcessingandControl31(2017)127–138 137 Inthisstudy,subjectswereonlytestedatsubthreshold
stim-ulation amplitudes. STIM artifact durations may be larger for suprathresholdstimulationamplitudes.Largerstimulation ampli-tudesmayresultinlargerCI artifactamplitudes.Assumingthat the decay constant doesnot change,it takes longer for larger CIartifact amplitudestodecaybelow thenoise level.However, 11subjectsweretested,wherethestimulationamplitudesused werejustbelowthesubject’sbehavioralTlevels.TherangeofT levelsobservedinthesesubjectsisquitediverse,resultingin max-imumstimulationpulseamplitudesusedbetween108and190cu, andbetween86 and167cuforstimulationat500and900pps, respectively.Wewouldarguethattheresultsfromthisstudyare representative,sinceawidevarietyofstimulationlevelswasused. Hereonlyonestimulationelectrodeinthemiddleofthearray wasused.Thestimulationelectrodeisnotexpectedtohavemuch influenceontheCIartifactcharacteristics.Futureresearchshould focus on the influence of stimulation electrode position on CI artifactcharacteristics,whichcanbeevaluatedwiththetools pre-sentedinthisstudy.
5. Conclusion
Inmostsubjects,theCIartifactwasatleastpartlycausedby theSTIMartifact.BasedonthedatapresentedinFigs.5and9–11, itisnotrecommendedtouseaveragereferenceforEASSR mea-surements.CIartifactAGFslopesandinterceptsandSTIMartifact durationsarelargerinthecontralateralhemispherefortheaverage referenceconfigurationthanCzorFpzreference.Inthe contralat-eralhemisphere,thereferenceelectrodehasasignificantinfluence ontheCIartifactAGFslopeandinterceptforstimulationat500pps andontheSTIMartifactduration.Inthecontralateralhemisphere, smallervariabilitiesinCIartifactAGFslopes(at500pps)andSTIM artifactdurationswereobservedwhenmorefrontalreference elec-trodeswereused.STIMartifactdurationswerebetween0.7and 1.7msand0.7and2ms,atcontralateralandipsilateralrecording electrodes,respectively.Thisshouldmakeitpossibletoremove theCIartifactatthecontralateralrecordingelectrodeswitha lin-earinterpolationinmostsubjects,forstimulationat500pps.For stimulationat900ppsorforstimulationat500ppsatipsilateral recordingelectrodes,moreadvancedCIartifactattenuation meth-odsareneeded.
Acknowledgments
Theauthorswouldliketothankallsubjectswhoparticipated inthestudy.Furthermore,aspecialthanksgoesouttoRobertLuke andTomFrancartfortheirvaluablefeedbackthroughoutthestudy. ThisresearchworkwascarriedoutintheframeofResearchProject FWOnr.G.066213‘Objectivemappingofcochlearimplants’,IWT Project(IWT,110722)‘Signalprocessingandautomaticfittingfor nextgenerationcochlearimplants’andKULeuvenResearchCouncil CoEPFV/10/002(OPTEC).Thesecondauthorissupportedwitha Ph.D.grantbytheHermesfonds(141243).Noneoftheauthorshave potentialconflictsofinteresttobedisclosed.
References
[1]J.C.Lilly,J.R.Hughes,E.C.Alvord,T.W.Galkin,Brief,noninjuriouselectric waveformforstimulationofthebrain,Science121(1955)468–469.
[2]T.Boons,J.P.L.Brokx,I.Dhooge,J.H.M.Frijns,L.Peeraer,A.Vermeulen,J. Wouters,A.vanWieringen,Predictorsofspokenlanguagedevelopment followingpediatriccochlearimplantation,EarHear.33(5)(2012) 617–639.
[3]C.J.Brown,M.L.Hughes,B.Luk,P.J.Abbas,A.Wolaver,J.Gervais,The relationshipbetweenEAPandEABRthresholdsandlevelsusedtoprogram thenucleus24speechprocessor:datafromadults,EarHear.21(2)(2000) 151–163.
[4]D.CafarelliDees,N.Dillier,W.K.Lai,E.vonWallenberg,B.vanDijk,F.Akdas, M.Aksit,C.Batman,A.Beynon,S.Burdo,J.-M.Chanal,etal.,Normative findingsofelectricallyevokedcompoundactionpotentialmeasurements usingtheneuralresponsetelemetryoftheNucleusCI24Mcochlearimplant system,Audiol.Neurotol.10(2)(2005)105–116.
[5]M.L.Hughes,C.J.Brown,P.J.Abbas,A.A.Wolaver,J.P.Gervais,Comparisonof EAPthresholdswithMAPlevelsintheNucleus24cochlearimplant:data fromchildren,EarHear.21(2)(2000)164–174.
[6]C.A.Miller,N.Hu,F.Zhang,B.K.Robinson,P.J.Abbas,Changesacrosstimein thetemporalresponsesofauditorynervefibersstimulatedbyelectricpulse trains,J.Assoc.Res.Otolaryngol.9(1)(2008)122–137.
[7]A.T.Herdman,O.Lins,P.VanRoon,D.R.Stapells,M.Scherg,T.W.Picton, Intracerebralsourcesofhumanauditorysteady-stateresponses,Brain Topogr.15(2)(2002)69–86.
[8]T.W.Picton,M.S.John,A.Dimitrijevic,D.Purcell,Humanauditory steady-stateresponses,Int.J.Audiol.42(4)(2003)177–219.
[9]R.A.Dobie,M.J.Wilson,Acomparisonofttest,Ftest,andcoherencemethods ofdetectingsteady-stateauditory-evokedpotentials,distortion-product otoacousticemissions,orothersinusoids,J.Acoust.Soc.Am.100(4)(1996) 54–71.
[10]M.Hofmann,J.Wouters,Electricallyevokedauditorysteadystateresponses incochlearimplantusers,J.Assoc.Res.Otolaryngol.11(2)(2010)267–282.
[11]X.Li,K.Nie,F.Karp,K.L.Tremblay,J.T.Rubinstein,Characteristicsofstimulus artifactsinEEGrecordingsinducedbyelectricalstimulationofcochlear implants,in:3rdInternationalConferenceonBiomedicalEngineeringand Informatics,2010,pp.799–803.
[12]M.Hofmann,J.Wouters,Improvedelectricallyevokedauditorysteady-state responsethresholdsinhumans,J.Assoc.Res.Otolaryngol.13(4)(2012) 573–589.
[13]R.J.Ilmoniemi,J.C.Hernandez-pavon,N.N.Mäkelä,J.Metsomaa,T.P.Mutanen, J.Sarvas,DealingwithartifactsinTMS-evokedEEG,in:AnnualInternational ConferenceoftheIEEEEngineeringinMedicineandBiologySociety,2015, pp.230–233.
[14]P.M.Gilley,A.Sharma,M.Dorman,C.C.Finley,A.S.Panch,K.Martin, Minimizationofcochlearimplantstimulusartifactincorticalauditoryevoked potentials,Clin.Neurophysiol.117(8)(2006)1772–1782.
[15]F.-C.Jeng,P.J.Abbas,C.J.Brown,C.A.Miller,K.V.Nourski,B.K.Robinson, Electricallyevokedauditorysteady-stateresponsesinGuineapigs,Audiol. Neurootol.12(2)(2007)101–112.
[16]B.A.Martin,Cantheacousticchangecomplexberecordedinanindividual withacochlearimplant?Separatingneuralresponsesfromcochlearimplant artifact,J.Am.Acad.Audiol.18(2)(2007)126–140.
[17]D.P.Allen,E.L.Stegemöller,C.Zadikoff,J.M.Rosenow,C.D.MacKinnon, Suppressionofdeepbrainstimulationartifactsfromthe
electroencephalogrambyfrequency-domainHampelfiltering,Clin. Neurophysiol.121(8)(2010)1227–1232.
[18]A.Santillán-Guzmán,U.Heute,M.Muthuraman,U.Stephani,A.Galka,DBS artifactsuppressionusingatime–frequencydomainfilter,in:Annual InternationalConferenceoftheIEEEEngineeringinMedicineandBiology Society,2013,pp.4815–4818.
[19]D.Sinkiewicz,L.Friesen,B.Ghoraani,Analysisofcochlearimplantartifact removaltechniquesusingthecontinuouswavelettransform,in:Annual InternationalConferenceoftheIEEEEngineeringinMedicineandBiology Society,2014,pp.5482–5485.
[20]M.Yochum,S.Binczak,Awaveletbasedmethodforelectricalstimulation artifactsremovalinelectromyogram,Biomed.SignalProcess.Control22 (2015)1–10.
[21]B.H.Boudreau,K.Englehart,A.D.C.Chan,P.Parker,Subthresholdtraining:a novelapproachtostimulusartifactcancellationinsomatosensoryevoked potentialrecordings,in:25thAnnualInternationalConferenceoftheIEEE EMBS,2003,pp.2659–2662.
[22]B.H.Boudreau,K.B.Englehart,a.D.C.Chan,P.a.Parker,Reductionofstimulus artifactinsomatosensoryevokedpotentials:segmentedversussubthreshold training,IEEETrans.Biomed.Eng.51(7)(2004)1187–1195.
[23]P.Grieve,P.Parker,B.Hudgins,K.Englehart,Nonlinearadaptivefilteringof stimulusartifact,IEEETrans.Biomed.Eng.47(3)(2000)389–395.
[24]S.Kohli,A.J.Casson,RemovaloftranscranialACcurrentstimulationartifact fromsimultaneousEEGrecordingsbysuperpositionofmovingaverages,in: AnnualInternationalConferenceoftheIEEEEngineeringinMedicineand BiologySociety,2015,pp.3436–3439.
[25]M.Mancini,M.C.Pellicciari,D.Brignani,P.Mauri,C.DeMarchis,C.Miniussi,S. Conforto,AutomaticartifactsuppressioninsimultaneoustDCS-EEGusing adaptivefiltering,in:AnnualInternationalConferenceoftheIEEEEngineering inMedicineandBiologySociety,2015,pp.2729–2732.
[26]V.Parsa,P.Parker,R.Scott,N.Brunswick,Convergencecharacteristicsoftwo algorithmsinnon-linearstimulusartefactcancellationforelectricallyevoked potentialenhancement,Med.Biol.Eng.Comput.36(2)(1998)202–214.
[27]L.M.Friesen,T.W.Picton,Amethodforremovingcochlearimplantartifact, Hear.Res.259(2010)95–106.
[28]M.McLaughlin,A.L.Valdes,R.B.Reilly,F.-G.Zeng,Cochlearimplantartifact attenuationinlateauditoryevokedpotentials.Asinglechannelapproach, Hear.Res.302(2013)84–95.
[29]D.A.Wagenaar,Real-timemulti-channelstimulusartifactsuppressionby localcurvefitting,J.Neurosci.Methods120(2)(2002)113–120.
[30]T.Wichmann,Adigitalaveragingmethodforremovalofstimulusartifactsin neurophysiologicexperiments,J.Neurosci.Methods98(1)(2000)57–62.
[31]L.F.Heffer,J.B.Fallon,Anovelstimulusartifactremovaltechniquefor high-rateelectricalstimulation,J.Neurosci.Methods170(2)(2008)277–284.
[32]U.Hoffmann,W.Cho,A.RamosMurguialday,T.Keller,Detectionandremoval ofstimulationartifactsinelectroencephalogramrecordings,in:Annual InternationalConferenceoftheIEEEEngineeringinMedicineandBiology Society,EMBS,2011,pp.7159–7162.
[33]D.D.E.Wong,K.A.Gordon,Beamformersuppressionofcochlearimplant artifactsinanelectroencephalographydataset,IEEETrans.Biomed.Eng.56 (12)(2009)2851–2857.
[34]I.Akhoun,C.M.McKay,W.El-deredy,Electricallyevokedcompoundaction potentialartifactrejectionbyindependentcomponentanalysis:technique validation,Hear.Res.302(2013)60–73.
[35]J.C.Hernandez-Pavon,J.Metsomaa,T.Mutanen,M.Stenroos,H.Mäki,R.J. Ilmoniemi,J.Sarvas,Uncoveringneuralindependentcomponentsfromhighly artifactualTMS-evokedEEGdata,J.Neurosci.Methods209(1)(2012) 144–157.
[36]R.J.Korhonen,J.C.Hernandez-Pavon,J.Metsomaa,H.Mäki,R.J.Ilmoniemi,J. Sarvas,Removaloflargemuscleartifactsfromtranscranialmagnetic stimulation-evokedEEGbyindependentcomponentanalysis,Med.Biol.Eng. Comput.49(4)(2011)397–407.
[37]J.Metsomaa,J.Sarvas,R.J.Ilmoniemi,Multi-trialevokedEEGandindependent componentanalysis,J.Neurosci.Methods228(2)(2014)15–26.
[38]N.C.Rogasch,R.H.Thomson,F.Farzan,B.M.Fitzgibbon,N.W.Bailey,J.C. Hernandez-Pavon,Z.J.Daskalakis,P.B.Fitzgerald,Removingartefactsfrom TMS-EEGrecordingsusingindependentcomponentanalysis:importancefor assessingprefrontalandmotorcortexnetworkproperties,NeuroImage101 (2014)425–439.
[39]R.Ruach,R.Mitelman,E.Sherman,O.Cohen,Y.Prut,Anassumption-free quantificationofneuralresponsestoelectricalstimulations,J.Neurosci. Methods254(2015)10–17.
[40]F.C.Viola,M.DeVos,J.Hine,P.Sandmann,S.Bleeck,J.Eyles,S.Debener, Semi-automaticattenuationofcochlearimplantartifactsfortheevaluationof lateauditoryevokedpotentials,Hear.Res.284(2012)6–15.
[41]F.C.Viola,J.D.Thorne,S.Bleeck,J.Eyles,S.Debener,Uncoveringauditory evokedpotentialsfromcochlearimplantuserswithindependentcomponent analysis,Psychophysiology48(11)(2011)1470–1480.
[42]J.Wolfe,E.Schafer,ProgrammingCochlearImplants,Tech.Rep.,2010.
[43]H.H.Jasper,The10–20electrodesystemoftheInternationalFederation, Electroencephalogr.Clin.Neurophysiol.10(2)(1958)371–375.