Robotics combined with electrical stimulation: hybrid support of arm and hand for functional training after stroke

Robotics

combined with

Electrical Stimulation

hybrid support of arm and hand for

functional training after stroke

Ard Westerveld

ombined with

Ele tri al Stimulation

hybrid support of arm and hand for

fun tional training after stroke

PrintedbyGildeprintDrukkerijen

Roboti s ombinedwithele tri alstimulation

hybrid supportof armandhandforfun tionaltrainingafterstroke ArdWesterveld

ISBN978-90-365-3600-4 ©2014,ArdWesterveld

All rightsreserved. No part of thisbookmaybereprodu edin anyformwithout thewrittenpermissionoftheauthor.

HYBRIDSUPPORTOFARMANDHANDFORFUNCTIONALTRAININGAFTERSTROKE

PROEFSCHRIFT

terverkrijgingvan

de graadvando toraandeUniversiteitTwente opgezag vande re tormagni us,

prof. dr. H.Brinksma,

volgensbesluitvanhetCollegevoorPromoties inhetopenbaarteverdedigen

opdonderdag13maart2014om14.45uur

door

Ard Johan Westerveld

geborenop25juni1984 teGendringen

Prof. dr. ir. H.vander Kooij Prof. dr. ir. P.H.Veltink en doorde assistent-promotor: Dr. ir. A. C.S houten

ISBN978-90-365-3600-4 Copyright2014

1 Generalintrodu tion 1 2 Sele tivityand resolutionof surfa e ele tri alstimulation

forgraspandrelease 10

3 Controlof thumbfor eusing surfa e fun tionalele tri al

stimulationandmus leload sharing 28

4 Grasp ontrolinstrokepatientsusing fun tionalele tri al

stimulationandmodelpredi tive ontrol 54

5 A roboti end pointmanipulatorforrehabilitation

exer isesafter stroke 76

6 Passive rea handgraspwithfun tionalele tri al

stimulationandroboti armsupport 98

7 Generaldis ussion 114 Referen es 124 Summary 138 Samenvatting 142 Dankwoord 146 Curri ulumVitae 150 Publi ations 152

1 Strokestrikesallovertheworld. Asiftheintegrityofmymind/body

onne -tion had somehow be ome ompromised, a ordingto dr. Bolte Taylor (2009) whodes ribesherownstrokeasastepbystepdeteriorationoftheintri ate neu-rologi al ir uitry. Although ea h strokeis unique (see Box 1.1), the ommon partisa ompromisedoxygensupplyto ertainbrainregionsresultingin elldeath andlossoffun tion.

Asea hbrainregionisresponsibleforaspe i fun tion,theee tsofastroke are highly dependent on the lo ation and size of the region in whi h the stroke o urred. In the rst period after her stroke, dr. Taylor ould not understand language,read,write,walkortalk. Impairments ausedbystrokein lude ompro-mised ontralateral motor ontrol, mus le weakness, spasti ity, memory de its, lossof sensation,visualimpairmentsand ompromised bladderandbowel ontrol (RothandHarvey2002). Inadditiontothesephysi al impairments,astroke an alsoinuen epsy hologi alfun tionsand anleadtodepression,fearandanxiety.

Box1.1: Stroke

Strokes are either is hemi (about 80% of all strokes) or hemorrhagi (gure 1.1). An is hemi stroke is hara terized by obstru ted erebral bloodow. Eitherbythrombosis, em-bolism or la unes (Roth and Harvey 2002).

Figure 1.1: S hemati representation of is hemi (left) and hemorrhagi (right) strokes.

(Almost) complete recovery

Minor impairments

Moderate to severe impairments

Severe impairments

requiring nursing home

Death

Figure1.2: Prognosisafterstroke a

Hemorrhagi strokes are aused by rupture of a blood vessel (Donnan et al. 2008) either inside the brain (intra erebral hemorrhage) or in the spa earoundthebrain(Subara hnoid hemorrhage). Only10%ofthestroke vi timswillfullyre over,otherseither dieshortlyafterstrokeorhaveto ope withminortosevereimpairments,see gure1.2.

a

1

1.1 Inuen eofstrokeondailylife

Worldwide,every threese ondsa newstrokesurvivor(and his/herfamily) hasto opewithsomeofthefun tionalimpairmentsdes ribedabove. Imaginenotbeing able to ommuni ateor expressyour feelings, annotrememberthings fromyour life before the strokeor be ome dependent on others for daily movement tasks. A ording to ES Lawren e et al. (2001), 77.4% of a ute stroke patients have upperlimbmotorde itsand72.4%havelowerlimbmotorde its. Compromised human motor ontrol(Box1.2)willlead tovariouslimitationsduringa tivitiesof daily living,likeeating, drinkingandpersonalhygiene,anddiminishthepatient's independen y.

Inahealthy situation,wearenot ons iouslyinvolvedin movingour limbsor inopeningand losingourhands. Un ons iouslywepredi ttheweightofa upof oee andpi kitup tobringit toourmouth todrink. That is,if you like oee of ourse, otherwiseyou wouldprobablythink twi e. Many strokepatientshave toworkvery hardtomovetheirarm ontralateral tothebrainlesionin adesired way. Overtimethismayimproveduetothe ompensatorystrategies(Roby-Brami et al. 2003)or plasti ity ofthe brain, i.e. thebrain's ability torearrange andlet other regionstakeoverfun tionsfromlostandae tedregions(Johansson2000; Nudoetal.2001;Barsi etal.2008).

Box1.2: Corti almotor ontrol

Voluntarymovementsareinitiatedto a hieve a desired goal. The brain integrates sensory information from the body and it's environment to drive the appropriate mus les to a - omplish a ertain task. During the task,sensorysignalsfrommus lesand skin are fed ba k to the brain and usedto ontrolthemovement (Kan-del et al. 2000). Mus le a tivation isdrivenfromtheprimarymotor or-tex (M1). M1 is lassi ally divided insubse tionsresponsiblefordistin t body parts (Nudo et al. 2001) om-monlyreferred toas thehomun ulus (little man)as shown in gure 1.3. Corti aldrivefromM1isproje tedto thealpha motor neuronin the spinal ord through the orti ospinal tra t.

The orti ospinaltra t rossestothe oppositesideofthespinal ord: right sidedmovementsare ontrolledbythe left hemisphere and vi e versa. The nerveendingsofthealphamotor neu-ron in the spinal ord innervate the mus lestogeneratethedesired move-ment(Kandelet al.2000).

M1

Figure1.3: Anatomi aldivisionsofthe pri-marymotor ortex(Redrawnbasedon Pen-eldandRasmussen1950)

1 1.2 Motor(re)learning

Peoplelearntheir wholelife. The basisoflearningistheformationofnewneural pathwaysand modi ation of existing pathways. After stroke, patientshave to (partially)relearnmotor ontrol. Motorlearningisdes ribedbyBastian(2008)as theformationofanewmotorpatternthato ursvialong-termpra ti e(i.e. days, weeks,years). A on ept loselyrelated to motorlearning is motoradaptation, whi h des ribes the modi ation of a movement due to per eived errors. For instan e,adaptationtobeableto usea omputer-mouse settoadierent speed as one is used to. This adaptationpro ess anturn into a learned alibration for thenewenvironment. Inrehabilitation,patientswho anonlymove slowlyor ina uratedonotneedtolearnthemovementfroms rat hbutdoneedsubstantial re alibrationfortheiraltered neural ontrol(Bastian2008).

Integration of sensoryinformation is an important fa tor for (re)learning. In monkeyexperiments,inwhi htheprimarysensoryhandareawasablated,monkeys wereabletoperformpreviouslylearnedmovements,butwerenotabletolearnnew movements (Krakauer 2006). For generalization of tasks learned by training to tasks in daily life, repetitive training of the same movement seems insu ient. Whenpatientsareaskedtopi kupaglassatvariablepositions,theywillprobably learnthe movementof rea hing for a glass in a spe i pla eto a lesser extent, buttheymightbebetteringeneralizing thetasktoreallifeandalsoretentionof thelearned movement is expe ted to behigher in thevariablesetting (Krakauer 2006).

1.3 Therapyafterstroke

Somespontaneousre overy ano urafterstroke(Nudo2006). Tofurtherredu e impairment and enhan e fun tional independen e of stroke survivors, additional therapy is ommonly provided. Stroke therapy either exploits brain plasti ity to relearnmovementbyextensivetrainingorfo usesonstrategiesto ompensatefor lostfun tions. Appliedtrainingparadigmsin ludearmabilitytraining, onstraint-indu edmovementtherapy,bilateralarmtraining,fun tionalele tri alstimulation (box 1.3), intera tive robot therapy and virtual reality based therapy (Krakauer 2006; Timmermans et al. 2009). These therapies should fo us on task-oriented training(skilllearning) to obtain better generalizationfrom rehabilitationsetting todailylifea tivities(Timmermans etal.2009).

1.4 Fun tionalele tri alstimulation

The prin iples offun tionalele tri al stimulation (FES)areexplained inbox1.3. FESissu essfullyappliedasaprostheti systemtorepla elostfun tions,mainly after spinal ord injury (Sheer and Chae 2007; Snoek et al. 2000). FES an also beused as therapeuti system toimprove motorfun tion after stroke. FES training anin reasemus lestrengthandtherebyredu eweaknessduetonon-use (Powelletal.1999;Rosewilliametal.2012)and anredu epainand ontra tions

1

(Malhotra et al. 2012). In a systemati review of randomized lini al trials, de Kroon et al. (2002) identied positive trainingee ts of FEStraining on motor ontrol. Barsi etal.(2008)showedin reased orti alex itabilityafterpoststroke FEStraining,whi hindi atesregenerationofneuralpathways.

1.5 Rehabilitationroboti s

Robotsareinexhaustiveandthereforeanidealpartnerforintensiverepetitive fun -tionaltrainingafterstroke. Thepastde ades,severalroboti systemsforarmand handtherapyhavebeendesigned. MIT-manus(Hoganetal.1992),Hapti Master (Van der Linde et al. 2002), CADEN-7 (Perry et al. 2007), ARMin (Nef et al. 2007),Freebal(Stienenetal.2009b)andDampa e(Stienenetal.2009a)are

ex-Box1.3: Fun tionalEle tri alStimulation

Fun tional ele tri al stimulation (FES)evokesneurala tivityinmotor nervebers. Generateda tion poten-tials will lead to ontra tion of the mus le, seegure1.4. Animportant dieren e omparedtonormalneural a tivity is the reversed re ruitment order. With FES the thi kest motor nervebersarea tivatedrst,as op-posed to physiologi al a tivation in whi h the smallest-diameter nerves area tivatedrst(Sheer andChae 2007), leadingto more oarse move-ment and earlier fatigue. In addi-tion, to obtain smooth ontra tions with FES, motor units are a tivated syn hronouslywithrelativelyhigh fre-quen y,alsoleadingtorelativelyearly mus lefatigue.

Threetypesofele trodes anbeused totransfer thegeneratedstimulus to thenerve: 1) implanted,2) per uta-neous or 3) surfa e ele trodes. Im-plantedele trodeshavethebenetof properly ungtheele trodearound thenerveleadingtoverysele tive a -tivation. However, this highly

inva-sivesolutionismainlysuitablefor per-manent FES appli ations. Surfa e ele trodes are pla ed further from the target nerve and dedi ated ele -trode pla ementis requiredfor sele -tivemus lea tivation. However, ur-rent spreads outin thetissue under-neaththeele trodesanda tivationof multiplenerves annotalwaysbe pre-vented. Nevertheless,duetoits non-invasiveness, surfa e ele trodes are ommonlyusedinrehabilitation pra -ti e,espe iallyintrainingtherapy(de Kroonetal.2002).

Figure 1.4: S hemati overview ofmus le a tivationwithsurfa eFES(1) orinvasive alternatives: nerve u (2), intraspinal(3) or intra orti al (4) stimulation (Stein and Mushahwar2005)

1 amplesof eitherroboti exoskeletonsor end-point manipulatorsfor armtraining.

Alsosomesystemsforhand traininghavebeendeveloped (Worsnoppetal.2007; Lamber yetal.2007;Dovatetal.2008).

Twore entreviews evaluatedtheee tsof roboti stroketherapy(Prangeet al.2006;Krebsetal.2008). Theyboth on ludethatroboti therapy animprove motor ontrolofthe hemipareti upper limb. Roboti aidedtherapygives similar results as onventional therapy (Kwakkel et al. 2008) and roboti manipulators fa ilitatemore intensivetraining and obje tivemeasurements (Lum etal. 2002), without the need of a therapist being ontinuously present. Thus multiple pa-tients ouldtrainsimultaneouslyundersupervisionofasingletherapistorpatients mightevenuseroboti swithoutsupervisionathomeforintensivetrainingwiththe therapistonlymonitoringprogressregularly.

1.6 The MIAS-ATD proje t: a hybridapproa h

Roboti sisidealforintensiveandrepetitivetraining. However,fromame hani al pointofview,properlya tuatingthehandandngerswitharoboti devi ewithout interferingmovements isrelatively omplex. Fun tionalele tri al stimulationhas beensu essfullyusedfora tuationofhandandngersandmightthereforebean ex ellentextension for a roboti arm support system. A hybrid system willallow for assistan e of fun tional task-oriented movements, fo using on skill-learning andthereforehaspotentialasa rehabilitationdevi e, aimingatgeneralizationto a tivitiesofdailylife.

The ATD (A tive Therapeuti Devi e) bran h of the MIAS (Medi al Innova-tions for an Aging So iety) proje t fo uses on the development of a hybrid re-habilitation system. The proje t is a onsortiumof Dem on, ti Medizinte hnik, Use-Lab,RoessinghResear h&Development(RRD)andtheUniversityofTwente (UT) funded by Interreg IV-A, part of theEuropeanregional development fund. Within the onsortium requirements and possibilities for a hybrid rehabilitation system were analyzed. Prototype roboti s were built by Dem on and prototype stimulator equipment was provided by ti Medizinte hnik. The prototypes were evaluatedbyUse-Lab,RRDandUT.

1.7 Resear h questions

The main goal of this thesisis to develop andevaluate ontrolalgorithms for a hybridrehabilitation system ombining FESandroboti s. The thesiswillprovide answerstothefollowingquestionsthatariseforproper ontrolofahybrid rehabil-itationsystem.

ˆ Whi h mus les involved in grasp and release are available to target with surfa eFES?And towhatextent anthesemus lesbesele tivelya tivated withFES?

1

mus les? How anthisrelationbemodeledandusedto ontroltheredundant mus ularsystemwithFES?

ˆ CanthedevelopedprototypeFESsystema tivatehandmus lesproperlyfor fun tionalgraspandrelease?

ˆ Is the developed prototype roboti manipulator suitable for assistan e of fun tionalrea hmovements?

ˆ Isthehybridrehabilitationsystem ombiningroboti sforrea handFESfor graspandreleaseee tiveforpassivemovementsupport?

1.8 Thesis outline

In this thesis several experimental studiesare des ribed to answer the questions above and evaluate the prototype hybrid system. By the use of an automated system for stroke rehabilitation, whi h is also appli able in the patient's home, therapy anbeintensied. Ideally,anautomatedsystemshouldonlysupportwhen ne essary, thereby maximizing patient eort (Wolbre ht et al. 2008). However, in this thesisthe te hni al feasibility andperforman e is evaluatedand therefore the subje ts were asked to relax in thedes ribed experiments (i.e. no voluntary movement). Apassivesubje twillbethemostdemandingsituationforthesystem andisthereforeusedasevaluationsetting.

In Chapter 2 the possibilitiesfor sele tive a tivation of individual ngersby fun tional ele tri alstimulation areexplored. The main questionto beanswered iswhether itis possibletondspe i lo ationsforsele tivengermovementsin dierent healthysubje ts.

Chapter3usessele tivea tivationofthreethumbmus lesto ontrolthefor es generatedbythethumbintheplaneperpendi ulartothethumb. Amodelforthe relation between the stimulation parameters and the evoked for es is developed and evaluated in both healthy subje ts and stroke subje ts. Subsequently, the individualmus lemodelsareusedto ontrolthethumbfor e towardstargetfor e ve torsbysharingtheloadamongtheindividualmus les.

A shift towards position ontrol is made in Chapter 4, where the relation between mus le stimulation and nger movement is modeled and subsequently usedinamodelpredi tive ontroller. This ontrollerusestheestimatedmodeland predi tsthene essarystimulationparametersbasedondesiredngerjointangles. Toestimatetheperforman eofthis ontrolapproa h,realobje tsaregraspedand released inhealthysubje tsandstrokesubje ts.

InChapter5, thedesignandte hni alevaluationofanewa tive therapeuti devi e is presented. This roboti end point manipulator is apable of providing guidan efor esand ountera tingtheweightofthearmtomakearmmovements easier.

Chapter6 ombinesthesystemspresentedin hapter4and hapter5. The ombinationofroboti supportedrea hmovementandsupportofgraspandrelease

1 by fun tional ele tri al stimulation is evaluated during passive rea h, grasp and

releasetasks inhealthysubje tsandstrokesubje ts.

Finally, inChapter 7theresultsof thisthesisaresummarizedanddis ussed. Thedis ussionfo useson lini alimpli ationsoftheknowledge urrentlyobtained andtherequiredfuture stepstotranslatethisknowledgeto lini alappli ations.

surfa e ele tri al stimulation

for grasp and release

Publishedas: Westerveld,AJ,ACS houten,PHVeltink,andHvanderKooij(2012). Sele tivityand

resolutionofsurfa eele tri alstimulationforgraspandrelease. IEEETransa tionsonNeural SystemsandRehabilitationEngineering. 20(1),pp.94101

2

Abstra t

Ele tri alstimulationofarmandhandmus les anbeafun tionaltool forpatientswith motor dysfun tion. Su ient stimulationof ngerand thumb mus ulature an support naturalgrasping fun tion. Yet itremainsun lear howdierent grasping movements an besele tivelysupportedbyele tri alstimulation. Thegoalofthisstudyistodetermineto whatextenta tivationofindividualngersispossiblewithsurfa eele tri alstimulationfor thepurposeofrehabilitationfollowingstroke.

Theextensor digitorum ommunis(EDC)mus le,exorpolli islongus(FPL)mus leand thethenarmus legroup, allinvolvedingraspand release,were sele tedfor stimulation. Theevokedfor esinindividualngersweremeasured.Stimulationthresholdsandsele tive rangeswere determinedfor ea hsubje t. Ele trode lo ationswherethehighestsele tive rangeo urredwere omparedbetweensubje tsandinuen esofdierentisometri wrist positionswereassessed.

Inallsubje tssele tivestimulationofmiddlengerextensionandthumbexionwas pos-sible. Inaddition, sele tivestimulationofindexandringngerextensionwaspossiblein most ases. In9out ofthe10EDCsubje tswewere abletostimulate3orall4ngers sele tively.However,largevariabilityinele trodelo ationsforhighsele tivitywasobserved betweenthesubje ts.

Withinthedesignsofgraspingprosthesesandgraspingrehabilitationdevi es,thevariability of ele trode lo ations shouldbetaken into a ount. Theresults of ourstudy fa ilitate theoptimizationofsu hdesignsandfavoradesignwhi hallowsindividualizedstimulation lo ations.

2 2.1 Introdu tion

Graspandreleaseofobje tsis animportantfun tionin daily life. Bothgrasping and releasing be omes di ult or even impossible for large numbers of patients fromseveralpathologies. Su ientele tri alstimulation(ES)ofngerexorand extensor mus les, together with the thumb mus ulature, anhelp these patients tobe ome morefun tionally independent(e.g. Shimadaet al.2003) andregain manualdexterity.

Besidesdire tlyprodu ingfun tionalhandmovement,ESisusedtotrain fun -tionalmovements in strokepatients(e.g. Barsiet al. 2008). Fortherapeuti ES surfa e stimulation is preferred above per utaneous stimulation, be ause of the non-invasive hara ter. During therapeuti trainingsessions, ES anassist fun -tional movements, leading to motor re-learning of these movements (Krakauer 2006). Espe ially ES in ombination with voluntary eort enhan es motor re-learning(DB Popovi¢etal.2009).

Redu ed mus le sele tivity, after stroke for example, leads to impaired ne motorskills (Lang andS hieber 2004). IfES anbe used tosele tively a tivate mus les,it ouldbeusedtotrainne motor ontrol. Smallele trodesareableto morepre iselytargetmus lesormus lepartsforsele tivea tivationthanarelarger ele trodes. Thispre isetargeting,however,isin reasinglyvulnerabletodeviations in ele trode lo ation. Therefore, ele trodes should bepositionedpre isely, whi h willbemoretime- onsuming omparedtolargerele trodes.

Themus lemotorpointpositionsrelativetotheskinareknowntovaryamong dierent subje ts(Nathan 1979; Nathan 1990) and might hange during move-ments of the mus le itself or during the movementof nearby mus les (Cameron etal.1999). Iftheinter-subje tvariationandthevariationduetomovementboth aresmall, a general lo ation may bedetermined, leading to near-optimal stimu-lation for most patients. However, if the inter-subje t variation is substantial or stimulation lo ations vary largely during movement, a sear h pro edure for the individualizedlo ation willbene essary. Arrayele trodes, overingthevariations (Popovi¢-Bijeli¢etal.2005;MLawren eetal.2008;DBPopovi¢andMBPopovi¢ 2009)togetherwithanonlineself-learningalgorithmforele trodesele tion ould beasolutionin that ase.

Numerous obje ts manipulated during daily life (e.g. oee ups, bottles, spoonsorpen ils),requiresu essfulmovementofthethumbtoformafun tional grip. In addition,some patientssuer from involuntarily enlargedexor a tivity, whi hhampersextensionofindividualngers(e.g. Langetal.2009)andtherefore thereleaseof obje ts. Also, ontrolled losing ofthehand bysele tiveexion of thengers be omes moredi ult. In the pin h grip for instan e it is important thattheother ngersdonotinterfere withthea tivengersperformingthegrip. Forrehabilitation,whereassistan eshouldbeappliedonlywhenneeded,sele tive ngerextension(to ountera tenlargedexora tivity)andthumboppositionare thefo uswhendevelopingee tivetoolsforrelearninggraspandreleasefun tions. Anatomi ally,theextensordigitorum ommunis(EDC)mus le onsistsof sev-eral parts a tuatingthe dierent ngers. These parts areinnervated by dierent

2

nerve bran hes. Thus, theoreti allyit should be possible tosele tively stimulate extensionof individualngers(Leijnse etal.2008). However, whenvoluntary ex-tending a single nger, some movement of other ngers an be observed (van Duinenetal.2009). Thisresultsfrombothbiome hani al ouplingand ombined neuromus ular ontrol(Lang andS hieber 2004). When ES is applied to indu e movementthese ouplings analsobeexpe ted.

In thepast, several neuroprostheti ES devi eshave been developed (Mi era etal.2010),in ludingtheBionessH200(formerlyNessHandmaster)(Hara2008), Bioni Glove (Pro hazka et al. 1997) and Me Fes (Thorsen et al. 1999). All of these devi essu essfully use surfa eESto trainor aida tivitiesofdaily life. In the Bioness H200, ele trodes are xed to the orthosis at appropriate positions. On e these positions are determined, donning and dong be omes quite easy. Problems with all of these devi es in lude: somewhat limited mus le sele tivity and omplexityinappli ationdue toproblems withele trodepositioning(Mi era etal.2010).

Kelleretal.(2006),assessedsele tivityofESappliedtothengerexors. They observed ouplingsbetweenthedierent ngersinallsubje ts. Theywereableto sele tively a tivate the middle and ring ngers in all subje ts, althoughthis was not expressed quantitatively. Nathan (1990)assessedthreshold urrent levels for bothtargeted andoverowmus lesin bipolarES. Overowtoother mus leswas observed during stimulation of several arm mus les. Dierentparts of the EDC mus le-forsele tivengerextension-were not onsidered.

The goalofthe urrent study isto determinethe sele tivityandinter-subje t variability of ES applied to threemus les involved in grasping and releasing ob-je ts: extensor digitorum ommunis(EDC), exor polli is longus (FPL) and the thenar mus le group. The main fun tions of these mus les are extension of the ngers,exionofthethumbandabdu tion/oppositionofthethumb,respe tively. Knowledge of the sele tivity and the variability will give insight in the a ura y needed for ele trodepla ement,whi hformsimportantinputtothedevelopment ofnewtherapeuti toolsusingES.The moresele tiveamus le anbea tivated, themorepossibilitiesfornemotor ontrolwillbe ome available.

2.2 Methods 2.2.1 Subje ts

Intotal19healthysubje tsparti ipatedinthisstudy,dividedovertwosubgroups. Group1(N=10;agerange23-27yr;5male)parti ipatedintheextensordigitorum ommunispart ofthe study andthegroup2(N=9; age range23-30yr; 6male) parti ipatedin thethumb mus ulature partof thestudy. All measurementswere performedon thelefthand. Subje tsgaveinformed onsentandtheexperiments were ondu tedina ordan ewiththeDe larationofHelsinki.

2

Figure2.1: S hemati overview of ustom-madesetupfor measurementofnger for es. The subje t'sve ngerswere strappedinpre-loadedwires. Smallload- ellsmeasuredwiretension andassu hngerfor e.

2.2.2 Experimentalsetup

A ustom-madesetupwasused, onsistingofanele tri alstimulatorandasetup for measurementofngerfor es.

Ele tri alstimulation

A battery-powered and urrent- ontrolled monophasi ele tri al stimulator with a peak amplitude of 13.5 mA was used. A ustom-built Matlab/Simulink (The Mathworksin .,Nati k,USA)interfa e ontrolledthestimulatorwirelesslythrough aBlueTooth onne tion. Anoval-shapedele trodeof6x4 mwasusedastheanode anda roundele trode,1.5 min diameter, wasused asthe athode. Ele trodes withsimilarsizeshowedgoodresultsonbothsele tivityand omfortinasimulation studybyKuhnetal.(2010).

For emeasurement

Tomeasurengerfor e,a ustom-madesetupwasbuilt,seegure2.1. Thissetup onsisted of an aluminum frame in whi h the lower left arm of the subje t was strapped just proximal to the elbow and wrist joints. The setup allowed several isometri positions. Thengerswere onstrainedbypre-loadedwires. Thetension in the wires was measured by LSB200 load ells (Futek, Irvine, USA), with a maximumfor e apa ityof45.3N.

2

1

2

3

4

B

A

C

D

F

E

A B C D E

1

2

3

3

2

1 A

B

C

Figure2.2: TodeterminethepositionoftheEDCandFPLgridpoints,smallroundlabelswere pla edrelativetobonylandmarks.Equidistantpointsforele trodepla ementweredrawnbetween theselabels. Forthethenarmus ulaturea

3

× 3

gridof1 mspa edwasdrawnonthethenar, relativetothemeta arpalboneofthethumb.

2.2.3 Experimentalproto ol Ele trodepla ement

The anodewaspla edontheposteriorsideofthelowerarm,justproximaltothe ulnarstyloidpro ess. Topositionthe athode attheFlexorPolli isLongus(FPL) mus leandtheExtensorDigitorumCommunis(EDC)mus le,aweb am(Philips, Eindhoven,TheNetherlands)wasaddedtothesetupforvirtual proje tionofgrid points. Inaddition,theweb amwasusedtotakepi turesoftheele trodelo ation, see gure2.2. Forea hsubje t,thegridpoints weres aleda ordingtothesize of the subje t'sarm, as thepoints were dened relative tobony landmarks. For athodepla ementonthethenarmus les,a

3

× 3

gridof1 mspa edpointswas drawnonthethenar.

Stimulationproto ol

Themus leswereele tri allystimulatedwithsinglepulsesof350

µ

s

width. Every se ond a stimulus wasapplied. For fun tionalmovementpulsetrains witha fre-quen yof12-50Hzareoftenusedinsteadofsinglepulses. We hosetousesingle pulses to beable to dire tly onne t the measuredfor e responseto theapplied stimulation pulse,withouttheneedoftakingthepulsehistoryintoa ount.

The stimulusamplitudestartedat2mAandwasin reasedby0.5mAattwo se ondintervals,untilthesubje treportedunbearabledis omfortorthemaximum amplitudeof13.5mAwasrea hed. Formostsubje ts,13.5mAwasstillbearable, buttheyreportedthattheintensitywasontheedgeofpainfulstimulation.

2 2.2.4 Re ordings

SensordatawasampliedbySG-3016IsolatedStrain GaugeInputModules (ICP-DAS, Taipei,Taiwan)anda quiredbyaUSB-6259dataa quisitionmodule (Na-tionalinstruments,Austin,USA) togetherwithap runninga ustom-built Mat-lab/Simulink(TheMathworksin ., Nati k,USA)interfa e. For e responses were measuredat1.6kHz.

2.2.5 Dataanalysis

For edatafromea h sensorwaspre-pro essed intwosteps: 1)a rstorder But-terworth highpass lter witha ut-o frequen yof 1 Hzwas applied toremove driftand2)a50mswindowmovingaveragelterwasappliedtoredu enoise. Sele tionof responsethresholds andsele tiveranges

Forea hindividualnger,theele trodelo ationwiththelowestresponsethreshold wasdetermined. Athresholdof 0.025Nwasusedto dis riminatebetweensensor noiseandana tualfor eresponse. Thesele tiverangewasdeterminedastherange between the responsethreshold of the spe i nger andthe response threshold of anyother nger. The sizeofthe sele tiverange gives information abouthow sele tively a single nger an be stimulated. See gure 2.3 for an example of determinationofresponsethresholdsandsele tiveranges.

Variationbetweensubje ts

Forea h subje ttheele trodelo ation(s)withthelowestresponsethresholdsfor a spe i nger wasdetermined. This anbe multiplegrid pointswhen multiple pointshavethesameresponsethreshold. Forea hsubje t

i

,amatrix

G

i

withthe samesize as theele trode gridis determined.

G

i

is oneat the lowest threshold lo ation(s) andzerootherwise. Finally, thenormalized relativeo urren e

G

was determinedforallsubje tstogetherbysummingall

G

i

'sanddivisionbythenumber ofsubje ts,

N

,asdes ribedin equation2.1.

G

=

∑

G

i

N

(2.1)

2.2.6 Inuen eofaltered isometri position

Fivedierentisometri positionsweretested,seetable2.1. Thresholdandsele tive rangeweredeterminedfortheindexnger(EDCstimulation)andthethumb(FPL stimulation). Thresholdlevelsandsizeofsele tiverangesofthedierentisometri positionswere omparedtotheneutralpositionusingpairedt-testswithBonferroni orre tionformultiple omparisons.

Chapter 2

0

0.5

1

1.5

Force [N]

0

10

Current [mA]

0

20

40

Time [s]

Index finger extension force

Middle finger extension force

Ring finger extension force

Little finger extension force

I

M

R

L

0

2

4

6

8

10

12

Threshold currents per finger

R

M

I

M

I

R

Threshold &

selective region

for middle finger

L

0

0.5

1

1.5

Force [N]

0

0.5

1

1.5

Force [N]

0

0.5

1

1.5

Force [N]

Current [mA]

Current [mA]

0

2

4

6

8

10

12

Figure2.3:Determinationofresponsethresholdsandsele tiverangeafterEDCstimulationforasinglegridlo ation.Responsethresholdwasdetermined

forea hnger(indi atedbythelabeledarrows) asthestimulationamplitudewheretheresultingfor eex eededathresholdof0.025N.Thesele tive

rangeforasinglegrid-pointwasdenedastheamplituderangewhereonlyonengerrespondedtothestimulation.Inthisspe i example,theresulting

sele tiverangeis3-6mA.

2 Table2.1: Testedisometri wristpositions

Position Flexion/extension Pro/supination

1 neutral neutral 2

45

o

extension neutral 3

45

o

exion neutral 4 neutral

90

o

pronation 5 neutral

90

o

supination 2.3 Results 2.3.1 Sele tiveness ofstimulation

Figures2.4 and2.5 show plotsof thesele tive rangesfor thedierent ngersof thedierentsubje ts. Forallsubje ts,itispossibletosele tivelystimulatemiddle ngerextension (gure2.4). Inaddition, for mostsubje ts,sele tive stimulation ispossiblefortheindexandringngers. Sele tivestimulationofthelittlengeris a hievedinonly4of10subje ts. Forthestimulationofthumbmovement(gure 2.5), all subje ts show the possibility for sele tive stimulation. Sele tive ranges varywiththeele trodelo ations.

Ingure2.6,boxplotsofthesele tiverangesizesareshownforthefourngers andthe thumb(bothFPL and thenar stimulation). For ea h subje t thelargest sele tive range for a spe i nger is sele ted (highest grey bar in ea h plot of gure2.4 and2.5). The sele tiverangesfor indexandmiddle ngersaresimilar. Ade rease insele tive range is observedfor thering andlittle ngers. Sele tive rangesforthethumb are omparabletothoseofindexandmiddlengers. 2.3.2 Variationof responsethresholds withrespe tto gridpoints

Ingure2.7thenormalizeddistributionoflowest-thresholdgrid-pointsa ross sub-je ts(seeEq. 2.1)isshownfortheEDCmus le(A-D),theFPLmus le(E)andthe thenar mus les(F). For thedierent ngers, lusteringof grid-points an be ob-served. Thusthelowest-thresholdpointsforthedierent ngerslie losetogether forthedierentsubje ts. However,therewasalargeoverlapbetweenthedierent ngers. In the thumb mus les, the points with the lowest threshold were more spreadoverthegrid. Thus theele trodelo ationwherethestimulationthreshold waslowestvariedgreatly betweendierent subje ts.

2.3.3 Inuen eofaltered isometri position

Figure2.8 showsrespon es ofthreshold amplitudeandsele tive rangeto altered isometri positions for the Index nger and the thumb. Distributions over the subje ts omparedtotheneutralpositionareshown.

Therewasalargevariationintheresponsesforthedierentsubje ts. Forboth EDC stimulation andFPL stimulation,no signi antsystemati hangein either threshold amplitudeor sele tiverange due to thealtered isometri positionswas observed.

2

ABCD

E F

1 2 3

₄

0

5

10

Ring

ABCD

E F

1 2 3

₄

0

5

10

Middle

Little

ABCD

E F

1 2 3

₄

0

5

10

Index

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

ABCD

E F

1 2 3

₄

0

5

10

I [mA

]

Figure 2.4: Thresholds andsele tive rangesfor thesubje ts (rows) of group1 (N=10). The olumns present responses of ea h nger to stimulation of theextensor digitorum ommunis mus leatsele tedele trodepositions(squares). Datarepresentedasexplainedingure2.3.

2

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

A B

C

1 2

₃

0

10

Thenar

FPL

I [mA

]

A B C

DE

1 2

₃

0

10

A B C

DE

1 2

₃

0

10

A B C

DE

1 2

₃

0

10

A B C

DE

1 2

₃

0

10

A B C

DE

1 2

₃

0

10

A B C

DE

1 2

₃

0

10

I [mA

]

I [mA

]

AB C

DE

1 2

₃

0

10

AB C

DE

1 2

₃

0

10

A B C

DE

2 3

0

10

1

I [mA

]

I [mA

]

I [mA

]

I [mA

]

I [mA

]

I [mA

]

Figure2.5:Thresholdsandsele tiverangesforthesubje tsofgroup2(N=9).Ea hrowrepresents asinglesubje t.The olumnspresentresponsesofthethumbtostimulationoftheexorpolli is longusmus le(rst olumn)andthethenarmus legroup(se ond olumn)atsele tedele trode positions. Dark greybarsdenote the sele tiverange, the orresponding squares illustratethe sele tedele trodeposition.

2

Index

Middle

Ring

Little

-4

-2

0

2

4

6

8

10

Finger

Siz

e of selec

tiv

e r

eg

ion [mA

]

Thumb

(Thenar)

Thumb

(FPL)

Figure2.6: Boxplotsofthemaximalsele tiverangesforea hofthengers(EDCstimulation) and thethumb (both FPLandthenar stimulation)overallsubje ts. Onea hbox,the entral markisthemediansele tiverange,theedgesoftheboxarethe25thand75thper entiles,the whiskersextendtothemostextremesele tiverangeswhi harenot onsideredasoutliers,and theoutliersareplottedindividually.

A B

C D

E F

1

2

3

4

0

0.5

1

A B

C D

E F

1

2

3

4

0

0.5

1

A B

C D

E F

1

2

3

4

0

0.5

1

A B

C D

E F

1

2

3

4

0

0.5

1

Middle

Index

Ring

Little

A.

B.

C.

D.

A B

C

D E

1

2

3

0

0.5

1

A

B

C

1

2

3

0

0.5

1

Thumb

Thumb

E.

F.

Figure 2.7: Normalized relativeo urren e (

G

)ofsubje t-dependent lowest-threshold-positions for(A-D)allngersbasedonEDCgridpoints,(E)FPLgridpointsand(F)Thenarmus legrid points

2

-10

-5

0

5

10

T

hr

eshold amplitude diff

er

enc

e [mA

]

Index Thumb

Index Thumb

Index

Thumb

Index

Thumb

-10

-5

0

5

10

S

elec

tiv

e r

ange diff

er

enc

e [mA

]

Flexion

Extension

Pronation

Supination

Figure2.8: Boxplots of dieren esinindexngerand thumb a tivationthreshold (top) and sele tiverangesize (bottom) withrespe ttoneutralposition for dierent isometri positions. Positions shownarepro/supination ombinedwith

45

°ofexion(Flexion)and

45

°ofextension (Extension)andneutralexion/extension ombinedwith

90

°ofpronation(Pronation)and

90

°of supination(Supination). Onea hbox,the entralmarkisthemedian,theedgesoftheboxare the25thand75thper entiles, thewhiskersextendtothemostextremedata points whi hare not onsideredasoutliers,andtheoutliersareplottedindividually.

2

2.4 Dis ussion

All subje tsshowed thepossibilitytosele tively stimulateindividual nger exten-sionandthumb exion. We were able tosele tively stimulatethe thumbinall 9 subje ts. In all subje tsof theEDC group, wewere able to sele tivelystimulate at least 2 ngers. In 9 out of the 10 EDC subje ts we were able to stimulate 3 or all 4ngers sele tively. However sele tive extension of the little nger was nota hievedin 6of10 subje ts. Theseresultsindi atethatsomene ontrolof the ngersmightbepossiblewith theuseof ES. The ele trodepositionsleading to eitherthelowestthresholdamplitudeor thelargest sele tiverangevaried sub-stantially between subje ts. Thus, althoughit is possibleto sele tivelystimulate dierentngers,theappli ationofthissele tivestimulationrequiresknowledgeof theindividual propertiesofthesubje t. Inaddition,pla ementofthestimulation ele trode at thelo ation withthe lowest response thresholddoes not ne essarily yieldthelargestsele tiverange. Therefore,the hoi eofele trodelo ationshould dependon the required sele tivity ofthe task. Assisting ylindri algrasp/release for instan e will require less sele tiveness than assisting the pin h grip or other more omplexmanualtasks.

2.4.1 Physiologi alaspe ts

The fa t that sele tivestimulation is a hieved, is likely the result of stimulation ofindividual mus lepartsthroughindividualnervebran hes. Leijnseetal.(2008) observedarrangementsofdierentEDC mus lebellies ommontodierent spe -imens. They observed the mus le part of the little nger was not onsistently separable from the ring nger part. In addition, the tendon of this mus le part inserts intobothringandlittle ngers. This ould explainthe fa tthatwewere unable to sele tively stimulate the little nger in 6 of 10 subje tsin the urrent study.

Therelativelysmallsele tiverangesoftheEDCmus leobservedinour exper-iments might be aused by me hani al oupling of thetendons, by theso alled jun turae tendinum, whi h onne ts the tendons of the dierent ngers on the ba kof thehand (LangandS hieber2004). Inaddition, ouplingsina tive neu-romus ular ontrolmightinuen etheabilitytosele tivelya tivatea singledigit negatively. LangandS hieber(2004)observedthisneuromus ular ouplingtobe largestin the ontrolof ringandlittlengers,whi halso mighthave ontributed tothefa tthatwewereunabletosele tivelystimulatethelittlengerinourstudy. Mus lepositions relative to theskin hangeduring wrist movementand one wouldexpe tthatele tri alstimulationparametersvarywiththisposition hange. However,underalteredisometri positionswedidnotobservesystemati hangesin eitherthresholdlevelorsele tiverange. Ourobservationsdoindi atethatthereis alargevariabilitybetweenthesubje tsregardingtheinuen eofalteredisometri positions. Therefore, an individual approa h for identifying inuen e of altered wristpositionand ompensationforthepossiblyalteredresponseisdesirable.

2 2.4.2 Relatedwork

The urrentstudyshowedsimilarresultstothestudyofKelleretal.(2006). How-evertheylooked atthesele tivityof ngerexormus les,theyalsosu eeded in sele tivestimulation ofmostof thengers, butwere unabletosele tively stimu-latethelittlenger. Nathan(1990)didnotlookintothestimulationofindividual ngers, but was able to sele tively stimulate the thumb by the FPL mus le and thethenar mus ulature. FortheFPL thesele tiverangeswere quite similar. For the thenar mus ulature he observed mu h larger ranges. This might be aused by the usage of bipolar ele trodes instead of monopolar in our ase. In bipolar stimulation,the urrent anbetargetedmorepre ise. Thisislikelytohavemore ee tinsmallermus les,likethethenar mus ulature.

Re ently,Kuhnetal.(2009)showedthatbytheuseofaproper ombinationof gellayerresistivityanddistan ebetweentheele trodes, multipleele trodesinthe array anbeused toprodu ea largervirtual ele trode,with similarproperties of aphysi allylarger ele trode. They statethatthedistan ebetweentheele trodes shouldstaybelow 3mm tokeeplosses small. The larger thissize,thelargerthe gel layer resistivity needs to be. In another study, Kuhn et al. (2010) ompared stimulation omfortandstimulationsele tivity. Theresultsshowedthatthemost omfortableele trodesizedependsonthethi knessofthefatlayerandthedepth ofthenerve tobestimulated. Inthin fat layers andfor stimulationof super ial nerves, smaller ele trodes were more omfortable. Subje ts an tolerate higher urrentdensitiesonsmallerele trodes.

2.4.3 Limitations

For daily life appli ations,higher frequen ystimulation wouldbe moreuseful in-steadofsinglepulsestimulation,be ausehigherfor es anbeevoked. Thegoalof the urrentstudywastoassesstheextenttowhi hindividual ngers anbe a ti-vatedusingele tri alstimulation.Thisspatialsele tivitydependsonthegeometry oftheunderlyingtissues. Thegeometrymight hangeduetomovementofthewrist ordue to ontra tionofthemus le itself. Wedid notnd anysystemati ee ts ofdierentwristpositions. Inhigherfrequen ystimulationthe ontra tionofthe mus lewillbelarger omparedtosinglepulsestimulation. Therefore,the geome-try might hangemore. However, asour results indi atedierent wristpositions not having a systemati ee t on the sele tivity, we do not expe t mu h ee t of higher frequen ystimulation on the mus le sele tivity. Enoka and Fuglevand (2001) omparedtwit handtetanusdataofmus lesthat ontrolthedigitsofthe hand. Their omparisonalso indi ated thatfor these mus lesthe twit h-tetanus ratiodoes not hangesystemati ally within reasingfor e.

Wedidnotspe i allytargettheEDC mus le, butrathertargeted thedorsal skin of the proximal forearm under whi h the EDC is lo ated. As a result other nearbymus les, likeExtensorCarpi Ulnaris (ECU)andEDM,mightalso be stim-ulatedbythepulses. Sin ethewristwasxedinthesetup,a tivationoftheECU (awristmus le)shouldnotinuen eourresults. Aswewere unabletotargetthe

2

little nger sele tively in most ases, it is unlikely thatspe i a tivation of the EDMmus lehaso urredinsteadoftheEDCmus le.

Herehealthy subje ts were measured. In thefuture, this an be extended to subje tsfromdierent pathologiesmentionedbefore. Notethatmus ular proper-ties and(lo al) innervationof thearmmus les are notae tedasa dire tresult ofthementionedpathologies.Atalaterstage,duetoaltereduseorevennon-use, these properties will hange of ourse. But even after se ondary ompli ations geometry ofthe skin anditsunderlying mus les will not hangemu h. Itis this geometrywhi hisanimportantfa torforspatialsele tivityofsurfa estimulation. The observed inter-subje t variability in both ele trode position for sele tive stimulation andinuen es ofaltered isometri positionsina healthysubje t pop-ulationalreadydemandforanindividualizedapproa hforea h subje t. Although the responseof the plegi limbsof patients with neurologi aldamage is di ult to predi t, it is unlikely that variability will de rease. Thus, designs of future grasp-and-release rehabilitation devi es should in lude the possibility to position thestimulationele trodesa ordingtotheneedsoftheindividualpatient.

In the urrent study, wedid nottakeskin thi kness or thi knessof the sub- utaneous fat layers ofthe individualsubje tsinto a ount. Thisvariation infat layerthi knessmightexplain thevariabilityinstimulation levelsandsele tiveness partially,butitis expe tedthatthefat layersofour subje tshada mu hsmaller variabilitythanthevariabilityofthestimulation responses.

Subje t omfort was not expli itly measured in our study. Stimulation was stopped if subje ts reported unbearable dis omfort. Inmost ases subje tswere able to withstanda stimulation intensity of13.5 mA,whi h wasthe limit ofour stimulator hardware. In theory, stimulation hardwarewith a broader stimulation range, mighthave ledtodierentresults,i.e. largerstimulationranges. However, in most ases multiple ngers responded at a stimulation intensity of 13.5 mA, thus stimulation was not sele tiveanymore. In addition, mostsubje ts reported thestimulationintensityof13.5mAontheedgeofpainfulstimulation. Therefore, wedonotthinkthesomewhatsmallrangeofthestimulationhardwarehaslimited our results.

2.4.4 Impli ationsforrehabilitation

Wemeasuredisometri for esresultingfromsinglepulsestimulationtodetermine sele tivityof surfa eele tri al stimulation. Assu h we annotexa tly determine whetherthesele tivestimulationisappli ableinarehabilitationsettingorindaily life. However, we anrelatemeasuredfor es tothethumbfor e neededin lifting aglasslledwithwater(

≈

0.25kg)andngerfor esneededtoover omeenlarged a tivityofexormus les.

Liftinga0.25kgobje t,assuminga oe ientoffri tionof0.5,requiresafor e of 5N exerted by all ngers together. Kamikawa and Maeno (2008)estimated for e distribution ratios a ross the ngers and their phalanges: a required for e of 5 N leads to a desired for e of approximately 0.55 N to be exerted by the proximalphalanxofthethumb. EnokaandFuglevand(2001)estimateda

twit h-2 tetanusratio of 1:3 for the mus les ontrolling thedigits. Applyingthis ratioto

themaximumsele tivefor es urrentlymeasured,leadstoexertedfor esof0.6N attheproximalthumb phalanxduetotetani stimulation,whi hisenoughtolift a0.25kgobje t.

At the medial phalanges of the ngers we measured extension for es around 0.1N. A ordingtoMonster andHChan(1977), therelaxedEDC mus le hasa twit h-tetanusratioofabout1:5. Thisratioleadstoanestimatedtetani for eof 0.5Natea hofthemedialngerphalanges. Tothebestofourknowledge,there existsno literature on exion for es of individualngers dueto enlargeda tivity. Webelieveanestimatedtetani for esof0.5N anbeusedfor(atleastassistan e of)extensionofanindividualngersueringfromenlargedexora tivity.

Basedon thesenumbers,itislikelythatthesele tivestimulation weobserved in ourmeasurements anbe usefulforappli ation in rehabilitationanddaily life. However, dire tmeasurementswouldgiveamore learview onthisaspe t. 2.5 Con lusion

The goalof the urrent study was to determine thesele tivity and inter-subje t variabilityof ES applied tomus les involvedin grasp andrelease. The results of thisstudy show thatit is possibleto sele tively stimulatea single nger in most subje ts. However, the extent of this sele tive stimulation is highly variable be-tween dierent ngers and between dierent subje ts. In addition, the possible grid points for thissele tive stimulation dier stronglybetween subje ts. In our opinion,arrayele trodes arevery useful forfuture designsof grasping prostheses andgraspingrehabilitationdevi es. The use ofarrayele trodesprovides the pos-sibility of automati ustomization. So ES, even for more sele tive stimulation withsmallerele trodes, anbeappliedinaplugandplaymanner. Be auseofthe possible hangeof ele trodelo ationsduringmovementandthetimevarian eof themus ularsystem,anonlineself-learningalgorithmwhi h ontinuouslyidenties thebest ele trode lo ationsfor thegiven taskunder the hanging ir umstan es anbe used. A model whi h mapsele trode lo ationsto produ ed ngerfor es under dierent anglesand subje t properties will be useful to predi t out omes. Su h model anbeused ina laterstageto ontrolES ofgrasp andreleaseinan e ient manner. The results presented here, fa ilitate the optimization of su h te hniquesandthedevelopmentoffutureESdevi esingeneral.

surfa e fun tional ele tri al

stimulation and mus le load sharing

Publishedas: Westerveld,AJ,ACS houten,PHVeltink,andHvanderKooij(2013).Controlofthumb

for eusingsurfa efun tionalele tri alstimulationandmus leloadsharing. Journalof NeuroEngineeringandRehabilitation10(1),p.104

3

Abstra t

Strokesurvivors oftenhave di ulties inmanipulating obje ts with theirae ted hand. Thumb ontrolplaysanimportantroleinobje tmanipulation. Surfa efun tionalele tri al stimulation(FES) anassistmovement.Weaimto ontrolthe2Dthumbfor ebypredi ting thesumofindividualmus lefor es,des ribedbyasigmoidalmus lere ruitment urveand asinglefor edire tion.

Fiveablebodiedsubje tsandvestrokesubje tswere strappedina ustombuiltsetup. Thefor esperpendi ulartothethumbinresponsetoFESappliedtothreethumbmus les weremeasured. Weevaluatedthefeasibilityofusingre ruitment urvebasedfor eve tor mapsinpredi tingoutputfor es. Inaddition,wedevelopeda losedloopfor e ontroller. Loadsharingbetweenthethreemus leswasusedtosolvetheredundan yproblemhaving threea tuatorsto ontrolfor esintwodimensions. Thethumbfor ewas ontrolledtowards targetfor esof

0

.5 N

and

1

.0 N

inmultipledire tionswithintheindividual'sthumbwork spa e. Hereby,thepossibilitiestousethesefor eve tormapsandtheloadsharingapproa h infeedforwardandfeedba kfor e ontrolwereexplored.

Thefor eve tor predi tionof theobtainedmodelhadsmall RMSerrors withrespe tto thea tual measured for e ve tors (

0

.22 ± 0.17 N

for the healthy subje ts;

0

.17 ± 0.13 N

forthestrokesubje ts). Thestrokesubje tsshowed alimitedwork rangeduetolimited for e produ tionof theindividual mus les. Performan e offeedforward ontrol without feedba k,wasbetterinhealthysubje tsthaninstrokesubje ts. However,whenfeedba k ontrol was added performan es were similar between the two groups. Feedba k for e ontrollead,espe ially forthestrokesubje ts, toa redu tioninstationaryerrors, whi h improvedperforman e.

Thumb mus leresponses toFES an bedes ribedby asingle for e dire tion anda sig-moidalre ruitment urve.For eindesireddire tion anbegeneratedthroughloadsharing amongredundantmus les. Thefor eve tormapsaresubje tspe i andalsosuitablein feedforwardandfeedba k ontroltakingtheindividual'savailableworkspa eintoa ount. Withfeedba k,morea urate ontrolofmus lefor e anbea hieved.

3 3.1 Introdu tion

Strokehasbe omeamajor auseofmorbidityandmortalityinthewesternworld. In iden eofstrokealsoin reasesinlessdeveloped ountriesasaresultof hanging life-styles(OvbiageleandNguyen-Huynh 2011). Greying of so ietyand improved health- are are likely to result in an in rease of strokesurvivors. Fun tional in-dependen e of stroke survivors is highly inuen ed by their ability to perform a su essfulgrasp. Inmanya tivitiesofdailyliving,likedrinkingoropeningadoor, graspandreleaseisanessentialpartoftherequiredmovement.

Fun tionalele tri alstimulation(FES)ofhandmus les anbehelpfultotrain grasp and release in stroke subje ts (Crago et al. 1991; DB Popovi¢ and MB Popovi¢ 2009; Mi era et al. 2010). Depending on the ability of the individual patient, the assistan e may be sele tively ( hapter 2) in reased or de reased in ordertomaximizethevoluntarya tivitywhi hisimportantinrelearningmovements (Wolbre htetal.2008).

Grasping omprises oordinatedngerandthumbmotionand ontrolledfor e exertionontheobje ttobeheld. Asmus lesinitiatehumanmovement,a urate ontrolofmus le for e isaprerequisiteformovement ontrol. Forgraspingtasks thengers anberegardedassingledegreeoffreedom(DoF) joints,sin e move-mentoftheindividualphalangesis oupledbe auseof theundera tuationofthe nger. Furthermore, rotationalong the exion-extension axis of thenger is by farthemostimportantmovementforgraspingandreleasingobje ts. Thethumb, however,requiresadierentapproa hasitmovesalongmultipleaxes. Controlling for e and movementof the thumb will be most hallengingand may serve as a model, whi h may be generalized/redu ed to the single DoF ase for the other ngers.

A healthy thumb is a tuated in several dire tions by nine mus les in total (Kaufman etal. 1999; Pearlman et al.2004). However, notall ninemus les an betargetedproperly withsurfa e FES.Mainly, be auseof overlying mus lesand nearby sensorynervesmakingstimulation un omfortable. Therefore, onlyasmall subsetof thumbmus lesis availablefor FESwithsurfa eele trodes. Thislimits the movements whi h an be ontrolled with FES. However, thumb movements relevantforgrasping(mainlyopposition)arefeasiblewithsurfa eele trodes.

For edistributionovermultiplemus lesis ommonlyappliedin biome hani al modeling, solvinga tuator redundan y problems for a given task(Happee 1994; Prilutskyand Zatsiorsky2002). This loadsharingapproa hmight also beuseful fora tivatingaredundantmus uloskeletalsystem. Inaddition,bysharingtheload over allavailable mus leswemaximize theavailablerange of for e. However, to ourknowledge,loadsharinghasnotbeenappliedtoexternala tivationofmus les with surfa e ele tri al stimulation. We will evaluate this possibility and expe t thisapproa htoresultina uratefor e ontrolwithafor edistributionoverthe individual mus les optimizedby minimizingthe sum of squared re ruitment over allmus les.

Re ently, Lujan and Crago (2009) measured thumb for es evoked by three thumbmus les inhealthysubje tsandonespinal ordinjuredpatient. Usingthe

3

measured for es they trained anarti ial neural network (ANN) for feedforward for e ontrol. Theyshowedgood ontroloftheisometri thumbfor ein2D.With the urrent study weaimata moretransparentapproa h: usinglinear ombina-tions of estimated mus le for e ve tors instead of using a bla k-box ANN. This approa hgivesusthebenetoflearningmoreoftheunderlyingphysiologi al sys-tem, by omparing ombinedmus le responses with individual mus le responses. In addition,it might allow fora more generallyappli ableapproa h, withoutthe needoftraininganANN.

The goal of the urrent study is twofold: 1) Is it possibleto des ribethumb mus le responses to FES by a sigmoidal mus le re ruitment urve and a single dire tionoffor e? Andifso,aretheseso alledmus lefor emapssubje tspe i , suitable for stroke subje ts and time-invariant? And 2) Are mus le for e maps suitablefor usein 2D thumbfor e ontrolwithFES applyingload sharing? And if so, is feedforward ontrol onlysu ient andis theapproa h alsosuitable for strokesubje ts?

3.2 Methods

We will introdu e the proposed generalized mus le for e model for thumbfor e ontrolandmus leloadsharingrst. Thereafterwewilldes ribetheexperimental evaluationofthismodelinbothhealthysubje tsandstrokesubje ts.

3.2.1 Generalizedmus lefor emodel

We aimed at predi ting mus le for e resulting from FES by a relatively simple model. At a spe i thumbposture weassumedthatthe for e dire tion ofea h mus le,

φ

i

,is onstant andthatanonlinearsigmoidalrelation existsbetweenthe stimulation amplitudeandthegeneratedmus lefor e.

|~F

i

(A

i

)| =

p

1i

1

+ e

−(Ai−p2i)

p3i

− C,

C

=

p

1i

1

+ e

p3i

p2i

(3.1)

InEq.3.1,

|~F

i

(A

i

)|

isthefor emagnitudeofmus le

i

atstimulusamplitude

A

i

;

p

i1

is related to thefor e saturationlevel, i.e. the maximal outputfor e of that mus le,

p

i2

isrelatedtotheine tionpointofthesigmoidalre ruitment urveand

p

i3

isrelatedtothehorizontals alingofthere ruitment urve,i.e. theamplitude range. The latter term in Eq. 3.1 is an oset term, ensuring zero for e if the amplitude is zero. The mus le for e dire tions,together with the maximalfor e amplitudesforea hmus lerepresentsthefor eve tormapforasystemofmultiple mus les,seegure3.1foranexample.

Feedforwardthumbfor emodel

We assumed a linear ve tor summation of the mus le for es a ting around the samejoint.

~

F

=

n

∑

i

=1

x

i

|~

F

max

,i

|

cos(

_sin

₍

_φ

φ

i

)

i

)



Control of thumb force using surface FES and muscle load sharing

3

Figure

3.

1:

An

example

of

the

fo

rce

vecto

r

ma

p

(direction

(left)

and

magnitude

(right)).

The

colo

red

lines

in

the

left

pane

sho

w

the

measurement

x-and

y-fo

rce

s

fo

r

the

ab

ducto

r

poll

icis

longus

(AbPL),

opp

onens

pollicis

(OpP)

and

the

e

xo

r

pollicis

brevis

(FPB)

muscles.

T

he

determined

muscle

fo

rce

directions

are

indicated

by

the

grey

lines.

The

small

va

riations

indic

ate

that

the

angles

are

relatively

constant

throughout

the

op

erating

range.

The

fo

rce

vecto

r

map

in

the

left

pane

is

sho

wn

on

top

of

an

overview

of

a

custom

built

setup

fo

r

restraining

wrist

movements

and

measurement

of

thumb

fo

rces

with

tw

o

pre-loaded

single

axis

fo

rce

senso

rs.

The

t

te

d

sigmoidal

recruitment

curves

fo

r

the

three

th

um

b

muscles

and

the

indivi

dua

lmeasurement

poi

nts

(steady

state

of

step

resp

onses

at

dierent

amplitudes)

are

sho

wn

on

the

right.

3

InEq.3.2,thepredi tedthumbfor eve tor

F

~

,istheve torsumofthe individ-ualmus le for es(

n

= 3

),modelled asa re ruitmentfra tion,

x

i

, ofthemaximal mus lefor emagnitudes,

|~

F

max

,i

|

.

The model ofEq. 3.2 wasused toobtain the mus le stimulationlevels given a desired thumb for e. This inverse problem isredundant: threemus les anbe stimulatedtoobtainathumbfor ein twodire tions. Inour(real-time) ontroller implementation,weaddressedthisredundan yproblembyminimizingthesquared mus le re ruitment. Minimal summed for e is a typi al riterion also used in mus uloskeletal modeling and load sharing studies (Happee 1994; Prilutsky and Zatsiorsky 2002). The re ruitment was modeled as a fra tion of the maximal for e,thusweobtainedaboundedproblemwhi h anbeformulatedasminimizing theve tornormshown:













F

max

~x − ~F

r













2

2

(3.3) Inwhi h

~

F

r

is the[2x1℄ olumnve torequaltothereferen efor eand

Fmax

is the[2x3℄matrix ontainingthemaximal

x

and

y

for esofea hofthethreemus les.

~x

isthe[3x1℄ olumnve torwithindividualmus lere ruitmentfra tions. Totake theboundson

x

intoa ountwereformulatedtheve tornormshownin3.3asthe equationshowninEq.3.4.

argmin

x

∈[0,1]

~x

T

F

_max

T

F

max

~x − 2~

F

r

T

F

max

~x + ~F

r

T

~

F

r

(3.4)

Sin ethelattertermisindependentof

x

,theoptimalre ruitment,

x

,minimizing Eq. 3.4 an bewritten as a quadrati problem of the form as shownin Eq.3.5, with

Q

= F

T

max

F

max

and

~c = F

T

max

~

F

r

.

argmin

x

∈[0,1]

1

2

~x

T

_Q

_{~x −~c}

T

_~x

(3.5) Finallythe al ulatedreferen efor esforea hmus le,

xF

max

,are onvertedto stimulationamplitudesbyusingtheinverse ofthesigmoidalre ruitment(Eq.3.1) urveshowninEq.3.6.

A

i

= −p

3i

ln



_p

1i

|~

F

i

| + C

− 1



+ p

2i

(3.6)

The ombinationof obtained stimulation amplitudes,

A

i

, is the ombination whi htheoreti allywouldprodu eafor eequaltothereferen efor e,

~

F

r

,oratleast thefor ewhi hisminimizingEq.3.3whenthesystemhasrea heditsboundaries ofoperation. The onstant

C

representstheosettermasintrodu edinEq.3.1. 3.2.2 Modelevaluation

Subje ts

Five able bodied subje ts (age 32

±

13 years, 3 men) and ve stroke subje ts (age 55

±

18, 4 men) were in luded for this study. Table 3.1 summarizes the

3 Table3.1:Strokesubje ts' hara teristi s

Subje t Age Sex Ae tedside Monthspost-stroke ARAT

S1 50 M L 44 52/57

S2 61 M R 156 3/57

S3 69 M L 45 24/57

S4 68 M L 46 17/57

S5 26 F L 58 2/57

ThemaximalobtainableA tionResear hArmtest(ARAT)s oreis57points(normalmovement).

hara teristi sfortheindividualstrokesubje ts. Thestudywasina ordan ewith thede larationofHelsinkiandwasapprovedbythelo almedi alethi s ommittee. Allsubje tsgave writteninformed onsent. Duringtheexperiments,the subje ts wereaskedtorelaxtheir mus les,inorder toavoidvoluntarymus lea tivation. Experimentalsetup

Either thedominantarm (healthysubje ts) or theae ted arm(strokesubje ts) wasstrappedina ustombuiltdevi e. Thissetupwasusedtoxatethewristand thehand inneutralpronosupination,andtomeasuretheisometri thumbfor e in twodire tionsperpendi ularto the axis ofthe thumb. For es were measuredby two45.3

N

load ells(Futek, Irvine)preloadedwithsprings. Seegure3.1.

Aspe ialbuilt3 hannelasyn hronousbiphasi ele tri alstimulator(TIC Medi-zin,Dorsten,Germany)wasusedtoapplytheele tri alstimulationpattern. Stim-ulationwasappliedata onstantfrequen y(30Hz)andpulsewidth(150

µ

s

). The amplitude ouldbe ontrolledvia ustombuilt ontrollerswithinthestimulator's range

[0 − 30mA]

instepsof

0

.125mA

. Asingle

50x50mm

anodewasusedtogether with

16x19mm

athodesforea h hannel. Ele trodeswithsimilarsizeshowedgood resultsonbothsele tivityand omfort inasimulationstudy(Kuhn etal.2010).

AnEtherCATI/Osystem(Be khoAutomationGmbH,Verl,Germany)using Matlab/xPC (The Mathworks, Natti k, USA) as EtherCAT master devi e was used to ontrol the stimulator parameters and to apture analog data from the for e sensors.

Experimentalproto ol Preparation

The Abdu tor polli is longus (AbPL), Opponens polli is (OpP) and Flexor pol-li is brevis (FPB) mus les were sele ted for stimulation. We expe ted to move thethumbsu iently in dire tionsneededfor graspand releasewiththese mus- les. OpP opposesthethumb (pre-grasp),FPBmovesthethumbinward(grasp) and AbPL moves the thumb up (release). Ele tri al stimulation was applied (

30Hz; 150

µ

s

)whenele trodeswere pla edinitially. Theamplitudewasin reased toevaluate responses andsubje t omfort. Ele trodeswere lo atedatthemotor pointsbasedonexplorationoftheresponsestoele tri alstimulation.Seegure3.2 for anexampleofele trodepla ement.

3

Figure3.2: Ele trodepla ement. Exampleofpla ementofele trodeon(top)AbPLand pla e-mentofanodeatthedorsumofthewristand(bottom)aboveFPBmus leandOpPmus le. The AbPLele trodeswaspla edjustmedialoftheradialbone,approximately5 mproximaltothe wristjoint,theOpPele trodewaspla edlaterallyonthethenar,about1/3ofthelengthofthe rstmeta arpalbone,measuredfromtheproximalside.TheFPBele trodewaspla edatabout halfthelength oftherstmeta arpalbone onthemedialsideof thethenar. Exa tele trode lo ationsweredeterminedexperimentallybasedonobservedresponsesandsubje t omfort.