Intention detection of gait initiation using EMG and kinematic data

(1)

Intention

detection

of

gait

initiation

using

EMG

and

kinematic

data

E.C.

Wentink

a,

*

,

S.I.

Beijen

a

,

H.J.

Hermens

a,b

,

J.S.

Rietman

b,c

,

P.H.

Veltink

a

a_Biomedical_Signals_and_Systems_group,_MIRA_Institute_for_Biomedical_Technology_and_Technical_Medicine,_University_of_Twente,_The_Netherlands b_Roessingh_Research_and_Development,_The_Netherlands

c

BiomedicalEngineering,MIRAInstituteforBiomedicalTechnologyandTechnicalMedicine,UniversityofTwente,TheNetherlands

1. Introduction

Gaitinitiationintransfemoralamputees(TFA)isdifferentfrom

non-amputees[1,2].In non-amputeesitconsistsoftwophases.

First,preparationsaremadeforthestepexecution[1,3,4].During

thisphaseposturaladjustmentsaremade,thecenterofpressure

movestowardstheleadinglimbandthebodyistiltedforward.

Subsequentlythecenterofpressuremovestowardsthetrailing

limbandthebodyistiltedfurtherforward.Thehipandkneeofthe

leadinglimbstarttoﬂexandtheanklestartstodorsiﬂextoprepare

fortoe-off,whichistheendoftheﬁrstphase.Inthesecondphase

thestepisexecuted.Itstartsattoe-offoftheleadinglimbandthe

bodyistiltedfurtherforward.Musclesinthetrailinglimbstabilize

thebody,duringswingoftheleadinglimb,andgeneratepush-off.

Theexecutionphaseendsatheel-strikeoftheleadinglimb[1–5].

InTFAthesetwophasesaresimilar,butthedurationdiffers

dependingonwhichlegisleading,theprostheticlegorthesound

leg.ItappearsthatTFAhavethetendencytostandontheirsound

leg for as longas possible and load theprosthesis as shortas

possible[1,2,5].

Artiﬁcial push-off of a transfemoral prosthesis during gait

initiationmaybebeneﬁcial,toallowamorenaturalprocessand

reduce effort needed from the sound leg [2]. However, gait

initiationmust bepredicted, becausetimingofpush-offis very

important.Push-offingaitisdescribedasthepartofthegaitcycle

whichbeginsatonsetofankleplantarﬂexionandendsattoe-off

[6].Startingpush-offtooearlywillpropeltheamputeebackwards.

Startingtoolatewilldissipatepush-offorevencauseastumble.To

providecontrolinputsforsupportedprostheticgaitinitiation,the

beginningandendoftheexecutionphase,toe-offandheel-strike

of theleading limb respectively, need to be detected for both

leading limb conditions. If in amputees the prosthetic leg is

leading,theprosthetickneeshouldﬂexattoe-offandbereadyto

taketheloadatheel-strike.Whentheprostheticlegistrailing,the

prosthesisshouldprovidepush-off[3,7].

Forthedetectionofgaitinitiationseveralsensorsmaybeused

like gyroscopesand accelerometers, but alsoelectromyography

(EMG)fromtheremainingmuscles.EMGofgaitinitiationin

non-amputeeswasmeasured inseveral studiesbutprimarilyatthe

lowerleg[3,8,9].EMGactivityinamputeesduringgaithasbeen

measured in a few studies and is comparable to that of

non-amputees[10–12].EMGduringgaitinitiationinTFAhasnotbeen

studiedpreviously.

Inertial sensors have frequently been used to assess gait.

However, few studieswerefound that used inertial sensorsto

assessgaitinitiation[13].

Most studies used a combination of an optical position

measurement system and force plates [1–3,5,8]. The authors

foundnostudiesonreal-timeintentiondetectionofgaitinitiation

in(non-)amputees.Wethereforestudiedgaitinitiationdetection

ARTICLE INFO

Articlehistory: Received9August2011

Receivedinrevisedform14March2012 Accepted13July2012 Keywords: Electromyography Transfemoral Intentiondetection Gaitinitiation ABSTRACT

Gaitinitiationintransfemoralamputees(TFA)isdifferentfromnon-amputees.

Thisismainlycausedbythelackofstabilityandpush-offfromtheprostheticleg.Addingcontroland artiﬁcialpush-offtotheprosthesismaythereforebebeneﬁcialtoTFA.

Inthisstudythefeasibilityofreal-timeintentiondetectionofgaitinitiationwasdeterminedby mimickingtheTFAsituationinnon-amputees.EMGandinertialsensordatawasmeasuredin10 non-amputees.OnlydataavailableinTFAwasusedtodetermineifgaitinitiationcanbepredictedintimeto controlatransfemoralprosthesistogeneratepush-offandstability.Toe-offandheel-strikeoftheleading limbareimportantparameterstobedetected,tocontrolaprosthesisandtotimepush-off.

Theresultsshowthattoe-offandheel-strikeoftheleadinglimbcanbedetectedusingEMGand kinematic data in non-amputees 130–260ms in advance. This leaves enough time to control a prosthesis.BasedontheseresultswehypothesizethatsimilarresultscanbefoundinTFA,allowingfor adequatecontrolofaprosthesisduringgaitinitiation.

ß2012ElsevierB.V.Allrightsreserved.

*Correspondingauthorat:BiomedicalSignalsandSystemsgroup,MIRAInstitute forBiomedicalTechnologyandTechnicalMedicine,UniversityofTwente,P.O.Box 217,7500AEEnschede,TheNetherlands.Tel.:+31534892766.

E-mailaddress:[email protected](E.C.Wentink).

ContentslistsavailableatSciVerseScienceDirect

Gait

&

Posture

j our na l ho me pa g e : w ww . e l se v i e r . com / l oca t e / ga i t po st

0966-6362/$–seefrontmatterß2012ElsevierB.V.Allrightsreserved.

(2)

innon-amputees,beforeadvancingtoTFA,butmimickingtheTFA

situation.WeuseddatawhichcanbemeasuredinTFA,i.e.upper

legmuscleactivityandinertialsensors.Thereforethedatacanbe

usedforupperlegprosthesis,lowerlegprosthesisorevenorthosis.

Intheseapplicationstheneedforstabilityandcontrolisdesiredin

ordertoimprovegaitinitiationandgait[1,2,13–15].

Thegoalofthisstudyistodetermineifgaitinitiationcanbe

detected from EMG of the upper leg muscles and/or inertial

sensors. Detection should be sufﬁciently early to eventually

support gait initiation in transfemoral prostheses users. The

currentstudywasperformedinnon-amputees.

2. Methods

2.1. Participants

Tenhealthyvolunteersparticipatedinthestudy,noneofthem

hadahistoryoflowerlimbinjuries,neurodegenerativediseasesor

anyskinconditions.Theexperimentswereapprovedbythelocal

EthicsCommitteeandaninformedconsentwasobtainedbefore

theexperiments.

2.2. Measurements

Kinematicdatawasmeasured(100Hz)using2inertialsensors

from Xsens (Enschede, Netherlands), with 3D accelerometers,

gyroscopesandmagnetometers.Electrodeswereplacedaccording

totheSENIAMstandards[16].Oneachmuscletwoselfadhesive

electrodes (Kendal, H93SG, Tyco healthcare, Germany) were

placedascloselytogetheraspossible.EMGmeasurementswere

performed with a 16 bipolar channel Porti-system from TMSi

(Oldenzaal,Netherlands)atasamplefrequencyof2048Hz,no

pre-ﬁlteringwasapplied.

Nine muscles were measured, due to a limited number of

availableEMGchannelsandtomimictheprostheticsituation.The

upperlegmusclesand inertial sensorswereplacedon oneleg,

whichisthe‘‘simulatedprostheticleg’’.Themeasuredupperleg

musclesare:them.gluteusmaximus(GMa),m.gluteusmedius

(GMe),m.tensor fasciaelatae(TFL),m. rectusfemorus(RF), m.

vastuslateralis(VL),m.bicepsfemoris(BF).Inﬁvesubjectsone

extramuscle,them.erectorspinea(ES),wasmeasured.

Atthelowerlegonthecontralateralsidethem.tibialisanterior

(TA), m. gastrocnemius medialis (GaM), m. soleus (Sol) were

measured,for referencepurposes.Thisis the‘‘simulatedsound

leg’’.Inﬁvesubjectsthesimulatedsoundlegwasthedominantleg

andintheotherﬁveitwasthenon-dominantleg.

Footswitches,placedmid-heelandattheﬁrstmetatarsalhead

of each foot, gave information about heel-strike and toe-off.

Subjectsworetheir ownlow-heeledshoes.Fig.1illustrates the

placement of the inertial sensors and EMG electrodes. To

synchronizeEMG,footswitchesandinertialsensorsa

synchroni-zationpulsewasgivenatthestartandendofeachmeasurement

whichwasvisibleinalldatasets.

2.3. Procedures

Forthegaitinitiationexperimentsthesubjectswererequiredto

standuprightwiththeirweightequallydistributedonbothfeet,

theinitialposture.Datarecording wasstarted. After 5s in the

initialposturethesubjectswereaskedtopressthe

synchroniza-tionbutton(sync)andstartwalking.After ﬁvepacestheywere

askedtostop,turnaround,returntotheinitialposture,wait2–3s,

pressthesyncandwalkback.Thiswasrepeatedfourtimeswithin

eachmeasurement.Twomeasurementswereperformedforeach

leadinglimbcondition,16trialspercondition.

In additiona postural swaymeasurement wasperformed, a

forwardandbackwardswayingmotion,withoutfallingforwardor

backward.Thiswasusedtocalculatethethresholdsfortheinertial

sensorsforgaitinitiationdetection.

2.3.1. Sensortobodycalibration

Theinertialsensorsexpresstheirdatainthesensorcoordinate

system(~s).Twocalibrationexerciseswereperformedtoconvert

this datatothebody coordinatesystem(~f),using therotation

matrix(Rfs)(~f¼Rfs~s).Inshortthecalibrationofthelowerlegwas

as follows.The subjectstoodupright, whereby thebody z-axis

equalsthegravity vectorwhich canbedescribed in thesensor

coordinatesystem.Subsequentlythesubjectsﬂexedthekneeﬁve

timestoabout908,wherethekneeisthebodyy-axis,allowingthis

axistobedeﬁnedinthesensorcoordinatesystem.Thex-axisis

subsequentlyobtainedbyacrossproductoftheyandzaxes.A

similarprocedurewasfollowedfortheupperlegsensor,usingthe

squatascalibrationexercise[17,18].Thisdatawassubsequently

low-passﬁlteredat10Hz.Finallytheaxiswithlargestamplitude

wasusedforfurtheranalysis.Fortheangularvelocitythiswasthe

bodyy-axis,fortheaccelerationitwasthebodyz-axis(seeFig.1).

2.4. Dataanalysis

EMGdataanalysiswasperformedintwoparts.Firstthelinear

envelopesoftheensembleaverageswerecalculated,todetermine

whichmusclesshowaclearchangeinactivitybeforetoe-off or

heel-strikeoftheleadinglimb.Secondly,fromtheselectedmuscles

theonsetoroffsettimingsweredetermined.

TheEMGdatawasﬁrsthigh-passﬁlteredusingasecondorder

Butterworth ﬁlter with a cut-off frequency of 20Hz [19]. To

calculatethelinearenvelopesthedatawassubsequentlyrectiﬁed

andlow-passﬁlteredwithasecondorderButterworthﬁlterat9Hz

[19].Tocalculatethetimings,thehigh-passﬁltereddatawas

low-passﬁlteredat500Hz[19].

Fig.1.Placementoftheinertialsensors(IS)andEMGelectrodesonthebody.One legsimulatestheprostheticleg(SPL,ingray).Atthislegalltheupperlegmuscles weremeasuredandtheinertialsensorswereplacedattheupperandlowerleg.At theotherleg,thesimulatedsoundleg(SSL),onlythelowerlegmuscleswere measuredforreference.

(3)

Onsetsof theselectedmuscles werecalculated using a

log-likelihood-ratiotest(AGLR),asdescribedbyStaude[20,21].This

algorithmissuitableforreal-timeEMGonsetoroffsetdetection.

Thewindow-sizeusedforthedetectionwas20ms,thethreshold

ofthealgorithmforon-offdetectionwassetat20[20–22].

Thedifferentphasesofgaitinitiationweredeterminedusing

thefootswitchdata.Push-offtakesplacebetweenheel-offofthe

trailinglimb,whichistheonsetofplantarﬂexion,andtoe-offofthe

trailinglimb[6].Forbothleadingandtrailinglimbheel-off,toe-off

andheel-strikeweredetermined.Thedataofeachmeasurement

wassubsequentlyseparatedintotrialsandthetrialswerealigned

at heel-strike of the leading limb. From the aligned trials the

ensembleaverageswerecalculatedpersubject.

TheintrasubjectvariabilityoftheEMGtrialswasdetermined

using the variance ratio for each subject and muscle for the

preparationandtheexecutionphase[8,23].Thevarianceratiois

thevarianceofthedatabetweengaitinitiationcyclesnormalized

to the total variance. The lower the score is, the higher the

repeatability.Differencesbetweenthepreparationandexecution

phasewereanalyzedusingapairedt-testwithap-valueof0.05

andBonferonnicorrections[8].

Postural sway measurements were performed to determine

thresholdsforthekinematicdataaftercalibration,todecideifthe

subjectwasperformingposturalswayorwasinitiatinggait.Ifthe

datafromthemeasurementsexceededtheswaythresholds,then

toe-offorheel-strikeoftheleadinglimbcouldbedetected.

3. Results

3.1. Ensembleaverages

Fig.2showsatypicalexampleoftheensembleaveragesofthe

upperlegmusclesandtheinertialsensors,ofonesubjectwhere

the simulated prosthetic leg was leading (left) and where the

simulatedprostheticlegwastrailing(right).Thearrowsshowthe

musclesandinertialsensorsthathaveconsistentactivitychanges

beforetoe-offorheel-strikeoftheleadinglimbforallsubjects.

Thevarianceratiosofthedifferentmusclesinthepreparation

andexecutionphaseareshowninFig.3(b).Theexecutionphaseis

signiﬁcantly better reproducible compared to the preparation

phaseincasetheprostheticlimbwasleading.Themusclesthat

canbeusedforthedetectionofgaitinitiationshowgenerallya

lower variance ratio than the other muscles, but this is not

signiﬁcant.Thestandarddeviations,thebetweensubject

varia-tion,arelargeinsomecasesbutdonotdifferbetweenthedifferent

conditions.

3.2. Timings

Heel-strikeoftheleadinglimbwasdetectedinallcasesandwas

usedasareferencemeasureforalltimings,becausetoe-offofthe

leading limb was not detected in all trials. This was due to

inaccurateormissingfoot-switchdata.Sometrialswereexcluded,

because subjects started walking with the wrong leg or no

detection took place at all. The number of subjects and the

percentageoftrialsincludedinthecalculationofthetimingsare

speciﬁedinTable2.

Results for the timings of thefootswitches can befound in

Table1.Totalpush-offtime(SD)was285ms(75),starting166ms

(66)beforeandending125ms(38)afterheel-strikeoftheleading

limb. Table2shows theonand offsettimings oftheupperleg

muscles before toe-off or heel-strikeof the leading limb in all

subjects.Table2alsoshowsthedetectionoftoe-offoftheleading

limbusinginertialsensordata,whichwasonlypossiblewhenthe

prostheticlegwasleading.

3.3. Simulatedprostheticlegleading

The TFL and the RF showed activity onset in the ensemble

averages,whichisconﬁrmedbytheaverageonsetof129–199ms

beforetoe-offoftheleadinglimb.TheVLandtheBFshowedonset

of activityabout 150ms before heel-strike.Accelerometer and

gyroscopedataexceededtheswaythresholds160–260msbefore

toe-off.Heel-strikecouldnotbepredictedfromthekinematicdata,

itcouldhoweverbedetected.

Fig.2.Muscleactivityoftheupperlegmusclesandinertialsensordataofthesimulatedprostheticlimb(SPL)duringgaitinitiation.Theensembleaverageistakenover16 trialsofonetypicalsubject.Thickblacklinesindicatetheaverageactivityandthegraysurfaceindicatesplusandminusonestandarddeviation.Theverticallinesindicatethe events:toe-offleadinglimb,heel-strikeleadinglimbandtoe-offtrailinglimb,respectively.Ontheleft,wheretheSPLwasleading,activitychangesareseenbeforetoe-offof theleadinglimbintheTFL,theRFandtheinertialsensordataandintheVLandBFbeforeheel-strikeoftheleadinglimb.Ontheright,theSPLwastrailingactivitychangesare seenintheGMe,theGMe,theTFLandtheBFbeforetoe-offoftheleadinglimb.TheGMaandESshowactivitychangesbeforeheel-strikeoftheleadinglimb.

(4)

3.4. Simulatedprostheticlegtrailing

Forthiscondition,theGMa,theGMe,theTFLandtheBFcould

predict toe-off of the leading prosthetic limb 200–224ms in

advance. Heel-strike of the leading limb was detected for this

conditionbytheGMa(offset)andtheES(onset)163–199msin

advance.Kinematicdatacouldnotbeusedtopredicttoe-offor

heel-strike.

4. Discussion

Thegoaloftheexperimentswastodetermineifgaitinitiation

canbe predicted in non-amputees using data which would be

availableinprostheticusersontheprostheticside,e.g.EMGand

kinematic data. EMG of the upper leg muscles shows distinct

patternsduringgait initiationand wassimilar tothat inother

studies[3,8,9].ForbothleadinglimbconditionsEMGoftheupper

legmusclesshowedactivitychanges130–220msbeforetoe-off

andheel-strike.TheRFandTFLcanbeusedforthepredictionof

toe-offandtheVLandBFforpredictionofheel-strikeoftheleading

(prosthetic)limb.TheGMa,theGMe,theTFLandtheBFcanpredict

toe-offandtheGMaandEScanpredictheel-strikeoftheleading

(sound)limb.Kinematicdatacouldpredicttoe-offoftheleading

(prosthetic)limb,158–260msinadvance.

4.1. Simulatedprostheticlegleading

PreviousstudiesshowedthatTFAhaveatendencytostartgait

initiation with the prosthetic limb, because fewer adjustment

strategies are needed to initiate gait withthe prosthesis [1,2].

Whentheprosthesisisleading,thekneeshouldﬂexattoe-offand

extendatheelstrike[3,7].Ashortpreparationandalongexecution

phasewereseeniftheprostheticlegwasleadingcomparedto

non-amputees[1].ButevenifthepreparationphaseinTFAishalfthe

Fig.3.Varianceratiosofallmusclesaveragedoverallsubjectswithonestandarddeviation.Thelowerthescoreis,thebetterthereproducibility.Thepreparationphaseshows asigniﬁcantlylowerreproducibilitythantheexecutionphasewhenthesimulatedprostheticlimbwasleading.(a)Thereproducibilitywithinsubjectswhenthesimulated prostheticlegwasleadingand(b)whenthesimulatedprostheticlegwastrailing.

Table1

Timingfootcontacts.

HOLL TOLL HOTL HSLL TOTL HSTL TOTL-HOTL HOLL-TOLL

Time(ms) 549 462 166 0 125 652 285 87

SD 49 49 66 38 18 75 61

Timingsdeterminedusingthefootswitchesaveragedoveralltrialsofallsubjects.TOLL,toe-offleadinglimb;HOLL,heel-offleadinglimb;HOTL,heel-offtrailinglimb;TOTL, toe-offtrailinglimb;HSLL,heel-strikeleadinglimb;HSTL,heel-striketrailinglimb.AminussignreferstotheeventtakingplacebeforeHSLL.Timingsareaveragedoverall subjects.

(5)

durationofthatinnon-amputeesthecurrentresultssuggestthere

is enough time to control the prosthesis. Due to the lower

reproducibilityofEMGinthepreparationphase,thepredictionof

toe-offneedstheinertialsensordata.Inmicroprocessorcontrolled

kneessomeoftheseinertialsensorsarealready buildin.

Heel-strikecanbepredictedusingtheEMGdata.

4.2. Simulatedprostheticlegtrailing

Ifinamputeesthesoundlegwasleadingthepreparationphase

waslongerandtheexecutionphaseshortercomparedtohealthy

individuals[1].Duetothelongerpreparationtheremaybemore

timetodetecttoe-offoftheleadinglimbinTFAcomparedto

non-amputees.Timingofpush-off,whentheprostheticlimbistrailing,

mayneedsomeconsideration.Toaddpush-offtoprostheticgait,

heel-offand toe-offof thetrailinglimbneedtobedetectedfor

correcttiming[3,6,7].However,ifnoankleﬂexiontakesplacein

theprosthesisheel-offandtoe-offwilloccuralmostatthesame

time.Detectionofheel-strikeandtoe-offoftheleadinglimbwillbe

moreuseful.Theresultsshowthatfourmusclesareabletopredict

toe-off of the leading (sound) limb with good reproducibility.

However,onlytheGMahasahighreproducibilityinheel-strike

detection. Heel-strike of the leading limb may not need to be

predicted (only detected) in this case, because push-off ends

125msafterheel-strike.

4.3. Methodicalconsiderations

Toe-offwasnotdetectedinalltrials,thefootswitchesdidnot

provideanyinformationabouttheappliedpressure.Furthermore,

attheinitialstancephaseofgaitinitiationtheweightofthesubject

shiftsbackwardsalittlewhichmayunloadthetoeswitchesofthe

leadinglimbandthereforeunloadstheswitchesbeforeactual

toe-off.

Theerectorspinaewasonlymeasuredinﬁvesubjects,during

the experiments we found that the erector spinae may give

valuableinformationonposturalchanges,thereforeitwasadded

later.DataoftheESmaybeusedfordetectionofheel-strikeifthe

prostheticlimbistrailing,butthevarianceratioswereamongthe

highest.Fortheﬁnalapplicationitisthereforenotuseful.

Inpreviousstudies,durationofactivityofsomemusclesinTFA

wasfoundtobealittlelongerthaninnon-amputees[10–12].This

shouldnotbeaproblemforgaitinitiationdetectioninTFA,aslong

asclearchangesinmuscleactivitycanbedetectedbeforetoe-off

andheel-strikeoftheleadinglimb.Foroffsetdetectionofamuscle

thismaymeanthatlesstimeisavailablepriortotheevent,butthis

wasonlyrelevantintheGMaifthesimulatedprostheticlegwas

trailing.Forlongerstumplengths,amputationatthedistalhalfof

theupperleg,allthesuggestedmusclesarelikelytobeavailableif

myodesis of myoplasty has been performed. For short stump

lengths, however, some of the suggested muscles may not be

availableanymore[10].

Although all data was processed in such a way that onset

detectioncanbeperformedreal-time,thereisneedforadecision

algorithm.Toimplementcontrolintoaprosthesis,similarresults

mustbefoundinTFAandmoreactivitiesshouldbeanalyzed.

4.4. Conclusions

DetectionofgaitinitiationfromEMGoftheupperlegmuscles

and kinematic data in simulated amputee gait initiation was

possible.Intentiondetectionofgaitinitiationallows130–260ms

forcontrolofaprosthesis.However,furtherstudiesareneededto

determinethepossibilitiestopredictgaitinitiationinTFA.

Acknowledgments

ThisresearchissupportedbytheDutchTechnologyFoundation

STW,whichispartoftheNetherlandsOrganizationforScientiﬁc

Research(NWO)andpartlyfundedbytheMinistryofEconomic

Affairs,AgricultureandInnovation,undergrantno.08003.

Conﬂictsofintereststatement

Authors statethat no conﬂictsofinterest arepresent inthe

research. References

[1]vanKeekenH,VrielingAH,HofAL,HalbertsmaJPK,SchoppenT,PostemaK, etal.Controllingpropulsiveforcesingaitinitiationintransfemoralamputees. JournalofBiomechanicalEngineering2008;130(011002):1–9.

[2]VrielingA,KeekenH,SchoppenT,OttenE,HalbertsmaJ,HofA,etal.Gait initiationinlowerlimbamputees.Gait&Posture2008;27:423–30. [3]MannR,HagyJ,WhiteV,LiddellD.Theinitiationofgait.JournalofBoneand

JointSurgery1979;61:232–9.

[4]WinterD.Thebiomechanicsandmotorcontrolofhumangait.Waterloo: UniversityofWaterlooPress;1987.

[5]MichelV,ChongR.Thestrategiestoregulateandtomodulatethepropulsive forces duringgait initiationin lowerlimbamputees. Experimental Brain Research2004;158:356–65.

[6]ErdemirA,PiazzaS.Rotationalfootplacementspeciﬁestheleverarmofthe groundreactionforceduringthepush-offphaseofwalkinginitiation.Gait& Posture2002;15:212–9.

Table2

Timingsoftheupperlegmuscles.

Leadinglimb Muscle On/off TOLL(SD)inms #Sub %Trials HSLL(SD)inms #Sub %Trials

SSL GMa On 220(97) 10 90 – – – GMe On 216(49) 10 87 – – – TFL On 224(62) 10 95 – – – BF On 200(89) 10 81 – – – GMa Off – – – 199(70) 10 78 ES On – – – 163(67) 5* 82 SPL TFL On 129(90) 10 82 – – – RF On 199(108) 10 82 – – – VL On – – – 145(71) 9 88 BF On – – – 155(45) 10 95

Sensor TOLL(SD)inms #Sub %Trials

Acc UL 232(34) 10 95

Acc LL 158(90) 10 95

Gyro UL 260(67) 10 95

Gyro LL 258(34) 10 95

Aminussignreferstotheeventtakingplacebeforetheevent.TOLL,toe-offleadinglimb;HSLL,heel-strikeleadinglimb;SPL,simulatedprostheticleg;SSL,simulatedsound leg;Acc,accelerometerdata;Gyro,gyroscopedata;UL,upperleg;LL,lowerleg;#,thenumberofsubjectsincluded,%,thepercentageoftrialsincluded,*,theESwasonly measuredin5subjects.

(6)

[7]RoseJ, Gamble J.Human walking.Philadelphia: LippincottWilliamsand Wilkins;2006.

[8]MickelboroughJ,vanderLindenM,TallisR,A.R.E.Muscleactivityduringgait initiationinnormalelderlypeople.Gait&Posture2004;19:50–7. [9]QueraltA,Valls-SoleJ,CastelloteJ.Speedingupgaitinitiationandgait-pattern

withastartlingstimulus.Gait&Posture2010;31:185–90.

[10]JaegersS,ArendzenJ,deJonghH.Anelectromyographicstudyofthehip muscles oftransfemoralamputees inwalking.Clinical Orthopaedicsand RelatedResearch1996;328:119–28.

[11]HuangH,KuikenTA,LipschutzRD.Astrategyforidentiﬁcationlocomotion modes usingsurface electromyography.IEEE TransactionsonBiomedical Engineering2009;56(1):65–73.

[12]HaK,VarolH,GoldfarbM.Volitionalcontrolofaprosthetickneeusingsurface electromyography. IEEE TransactionsonBiomedical Engineering2011;58 (1):144–51.

[13]NajaﬁB,MillerD,JarettB,WrobelJ.Doesfootweartypeimpactthenumber of stepsrequiredto reach gaitsteady state?:an innovative lookatthe impact of foot orthoses on gait initiation. Gait & Posture 2010;32: 29–33.

[14]TokunoC,EngJ.Gaitinitiationisdependentonthefunctionoftheparetic trailing limb in individuals with stroke. Gait & Posture 2006;24: 424–8.

[15]HansenA,MiffS,ChildressD,GardS,MeierM.Netexternalenergyofthe biologic and prosthetic ankle during gait initiation. Gait & Posture 2010;31:13–7.

[16]HermensH,FreriksB,MerlettiR,StegemanD,BlokJ,RauG,etal.European recommendationsforsurfaceelectromyography.Enschede:Roessingh Re-searchandDevelopment;1999.

[17]LuingeH,VeltinkPH,BatenC.Ambulatorymeasurementofarmorientation. JournalofBiomechanics2007;40:78–85.

[18]ZwartjesD,HeidaC,vanVugtJ,GeelenJ,VeltinkPH.Ambulatorymonitoringof activitiesandmotorsymptomsinParkinson’sdisease.IEEETransactionson BiomedicalEngineering2010;57(11):2778–86.

[19]MerlettiR,ParkerP.Electromyography.Hoboken:WileyInterscience;2004. [20]StaudeG, Wolf W.Objective motor responseonsetdetection in surface myoelectricsignals.MedicalEngineering&Physics1999;21(6–7):449–67. [21]StaudeG,FlacheneckerC,DaumerM,WolfW.Onsetdetectioninsurface

electromyographicsignals:asystematiccomparisonofmethods.In:EURASIPJ ApplSigProc2001,vol.1.2001.p.67–81.

[22]RoetenbergD, BuurkeJ, VeltinkP,FornerCorderoA,HermensH.Surface electromyographyanalysisforvariablegait.Gait&Posture2003;18(2):109–17. [23]HershlerC,MorrisM.Anoptimalitycriterionforprocessingelectromyography (emg)signalsrelatingtohumanlocomotion.IEEETransactionsonBiomedical Engineering1978;25(5):413–20.