<%BANNER%>
Personalized neuromusculoskeletal modeling to improve treatment of mobility impairments: a perspective from European res...
CITATION PDF VIEWER
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/AA00012404/00001
 Material Information
Title: Personalized neuromusculoskeletal modeling to improve treatment of mobility impairments: a perspective from European research sites
Series Title: Journal of NeuroEngineering and Rehabilitation
Physical Description: Book
Language: English
Creator: Fregly, Benjamin J.
Boninger, Michael L.
Reinkensmeyer, David J.
Publication Date: 2012
 Subjects
Subjects / Keywords: Musculoskeletal model
Neural control model
Orthopedic surgery
Neurorehabilitation
Biomechanics
 Notes
Abstract: Mobility impairments due to injury or disease have a significant impact on quality of life. Consequently, development of effective treatments to restore or replace lost function is an important societal challenge. In current clinical practice, a treatment plan is often selected from a standard menu of options rather than customized to the unique characteristics of the patient. Furthermore, the treatment selection process is normally based on subjective clinical experience rather than objective prediction of post-treatment function. The net result is treatment methods that are less effective than desired at restoring lost function. This paper discusses the possible use of personalized neuromusculoskeletal computer models to improve customization, objectivity, and ultimately effectiveness of treatments for mobility impairments. The discussion is based on information gathered from academic and industrial research sites throughout Europe, and both clinical and technical aspects of personalized neuromusculoskeletal modeling are explored. On the clinical front, we discuss the purpose and process of personalized neuromusculoskeletal modeling, the application of personalized models to clinical problems, and gaps in clinical application. On the technical front, we discuss current capabilities of personalized neuromusculoskeletal models along with technical gaps that limit future clinical application. We conclude by summarizing recommendations for future research efforts that would allow personalized neuromusculoskeletal models to make the greatest impact possible on treatment design for mobility impairments.
 Record Information
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution.
Resource Identifier: doi - 10.1186/1743-0003-9-18
System ID: AA00012404:00001

Downloads

This item is only available as the following downloads:

Article ( PDF )

Eprints Dublin Core METS ( XML )

BioMed Central XML


Full Text

PAGE 1

REVIEW OpenAccessPersonalizedneuromusculoskeletalmodelingto improvetreatmentofmobilityimpairments:a perspectivefromEuropeanresearchsitesBenjaminJFregly1*,MichaelLBoninger2andDavidJReinkensmeyer3AbstractMobilityimpairmentsduetoinjuryordiseasehaveasignificantimpactonqualityoflife.Consequently, developmentofeffectivetreatmentstorestoreorreplacelostfunctionisanimportantsocietalchallenge.In currentclinicalpractice,atreatmentplanisoftenselectedfromastandardmenuofoptionsratherthan customizedtotheuniquecharacteristicsofthepatient.Furthermore,thetreatmentselectionprocessisnormally basedonsubjectiveclinicalexperienceratherthanobjectivepredictionofpost-treatmentfunction.Thenetresultis treatmentmethodsthatarelesseffectivethandesiredatrestoringlostfunction.Thispaperdiscussesthepossible useofpersonalizedneuromusculoskeletalcomputermodelstoimprovecustomization,objectivity,andultimately effectivenessoftreatmentsformobilityimpairments.Thediscussionisbasedoninformationgatheredfrom academicandindustrialresearchsitesthroughoutEurope,andbothclinicalandtechnicalaspectsofpersonalized neuromusculoskeletalmodelingareexplored.Ontheclinicalfront,wediscussthepurposeandprocessof personalizedneuromusculoskeletalmodeling,theapplicationofpersonalizedmodelstoclinicalproblems,andgaps inclinicalapplication.Onthetechnicalfront,wediscusscurrentcapabilitiesofpersonalizedneuromusculoskeletal modelsalongwithtechnicalgapsthatlimitfutureclinicalapplication.Weconcludebysummarizing recommendationsforfutureresearcheffortsthatwouldallowpersonalizedneuromusculoskeletalmodelstomake thegreatestimpactpossibleontreatmentdesignformobilityimpairments. Keywords: Musculoskeletalmodel,Neuralcontrolmodel,Orthopedicsurgery,Neurorehabilitation,BiomechanicsIntroductionMobilityinvolveswalking,stairclimbing,posture,balance,manipulation,transfers,andotherlocomotion tasksandisthereforecentraltoqualifyoflife.Whenan individualincursamobilityimpairment,qualityoflifeis diminishedinproportiontotheextentoftheimpairment.Forexample,mildkneeosteoarthritiscanlimit participationindesiredrecreationalorathleticactivities withoutsignificantlyaffectingnormaldailyactivitiesand productivity.Incontrast,astrokecanmakeitnearly impossibletowalkormanipul ateobjects,significantly diminishinganindividual sabilitytobeselfsufficient andfunctioninsociety.Spinalcordinjurycanleavea personwithnormalupperextremityfunctionbutno remaininglowerextremityfunction,significantly impactingonlycertainaspectsofmobility. Treatmentsfordifferentmobilityimpairmentsare typicallystereotypical,w ithastandardmenuoftreatmentoptionsexistingforanyparticularmobilityimpairment.Forexample,severemedialcompartmentknee osteoarthritismaybetreatedsurgicallyusinghightibial osteotomy,unicondylarkneereplacement,ortotalknee replacement.Onceapatientseekssurgicaltreatmentfor debilitatingpainandsignificantlossoffunction,the clinicianmustchoosebetweenthesetreatmentoptions basedonclinicalassessmentofthepatient.Furthermore, theclinicianmustdeterminetheoptimalvaluesofthe parametersassociatedwithth eselectedtreatment(e.g., method,level,andamountofcorrectionfortibial osteotomy,andimplanttype,size,andpositioningfor jointreplacement).Asimilarsituationexistsfor *Correspondence:fregly@ufl.edu1DepartmentsofMechanical&AerospaceEngineering,Biomedical Engineering,andOrthopaedics&Rehabilitation,UniversityofFlorida,231 MAE-ABuilding,P.O.Box116250,Gainesville,FL32611-6250,USA FulllistofauthorinformationisavailableattheendofthearticleFregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 JNERJOURNAL OF NEUROENGINEERING AND REHABILITATION 2012Freglyetal;licenseeBioMedCentralLtd.ThisisanOpenAccessarticledistributedunderthetermsoftheCreativeCommons AttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse,distribution,andreproductionin anymedium,providedtheoriginalworkisproperlycited.

PAGE 2

rehabilitationandsurgical treatmentsofneurological disorderssuchasstroke,Parkinson sdisease,andcerebralpalsy.Inclinicalpractice,thefinaltreatmentplanis usuallyselectedbasedonsubjectiveclinicalexperience ratherthanonobjectivepredictionofpost-treatment functiondevelopedfrompatientdata. Personalizedcomputationalmodelsoftheneuromusculoskeletalsystemcouldfacilitateobjectiveprediction ofpatient-specificfunctionaloutcomefordifferenttreatmentdesignsbeingconsideredbytheclinician.Dependingontheintendedclinicalapplication,apersonalized neuromusculoskeletalmodelcouldaccountforpatientspecificanatomical(e.g.,skeletalstructureandmuscle linesofaction),physiological(e.g.,muscleforce-generatingproperties),and/orneurological(e.g.,constraintson achievablemuscleexcitationpatterns)characteristics,all withinthecontextofamultibodydynamicmodel.Personalizedmodelsfortreatmentdesignaremotivatedby thefactthatformanytreatments, onesizefitsnone. Everypatientisdifferentandpossessesuniqueanatomical,neurological,andfunctionalcharacteristicsthatmay significantlyimpactoptima ltreatmentofthepatient. Personalizedmodelsprovideonepossibleavenuefor increasedobjectivityintreatmentplanning,reducingthe likelihoodthatdifferentclinicianswillplandifferent treatmentsgiventhesamepatientdata.Ideally,virtual treatmentsperformedonapatient spersonalizedmodel wouldallowobjectiveandreliablepredictionofposttreatmentfunctionandthusidentificationofanoptimal treatmentplan.Suchpredictionswouldidentifynotonly thebesttypeoftreatment(includingpreviously unknowntreatments)butals otreatmentparametersto whichfunctionaloutcomeishighlysensitive(i.e.,which treatmentparametervaluesdoestheclinicianneedto getright"?). Thispaperexploreshowpersonalizedneuromusculoskeletalmodelscouldbeusedtoimprovetreatment designformobilityimpairments.Theexplorationis basedonasurveyofpersonalizedmodelingresearch beingperformedinEuropeandthusislimitedinits scope.ThesurveywasfundedbytheNationalScience Foundation(NSF)intheUnitedStateswiththegoalof synthesizingresearchrecommendationsandinforming researchfundingintheareaoftechnologytoimprove mobility.InOctoberof2010,twoteamsoffourpanelistsrecruitedbyNSFvisitedanumberofacademicand industrialsitesthroughoutEuropeoveraoneweektime period.Sincetimeandfinancialconstraintslimitedthe numberoflabsthatcouldbevisited,itwasnotpossible togatherinformationfromalllabsinEuropeperformingvaluableworkinthisarea.Giventhatthegoalofthe tourwastosurveythestate-of-the-artinEurope,we alsoomitdiscussionofvaluableworkbeingperformed bylabsoutsideofEurope.Theremainderofthispaper summarizesthepanel sfindingsrelatedtothepotential clinicaluseandbenefitofpersonalizedneuromusculoskeletalmodeling.ClinicalaspectsofpersonalizedmodelingInthissection,wediscusscurrentandfutureclinical usesofpersonalizedneuromusculoskeletalmodelsto designimprovedtreatmentsformobilityimpairments. Tosetthestage,webeginbydiscussingcommonreasonswhyhumanmovementdataarecollected,followed byaproposalforageneralprocesstofollowwhenusing personalizedmodelsinthet reatmentdesignprocess. Wethendiscussmobility-relatedclinicalproblemscurrentlybeingaddressedwithpersonalizedneuromusculoskeletalmodels,andweconcludethissectionby highlightinggapsinclinicalapplicationwherepersonalizedmodelscouldaddsignificantvalue.ClinicalpurposeofpersonalizedmodelingPre-treatmenthumanmovement(e.g.,motioncapture, groundreaction,muscleel ectromyographic,energy consumption),strength(e.g.,isometricandisokinetic dynamometer),andimaging(e.g.,magneticresonance (MR),computedtomography(CT),x-ray,fluoroscopic)dataprovidetheexperimentalmeasurements necessarytodevelopobjectivemodel-basedpredictionsofpost-treatmentfunction.AsdescribedbyDr. MariaGraziaBenedettiattheRizzoliOrthopedic InstituteinBologna,Italy,therearethreeprimaryreasonsforcollectinghumanmovementdatainaclinical setting: 1) Assessment -Assessaftertreatmenthowthetreatmentworkedforanindividualpatientoragroupof patients.Anexamplewouldbeusinggaitdatatoassess changesinwalkingspeedandkneeflexionanglefollowingtendontransferorlengtheningsurgeryinaspecific childorgroupofchildrenwithcerebralpalsy.Thisuse ofhumanmovementdataisrelativelycommon. 2) Identification -Identifyonanindividualpatient basiswhichpatientsshouldbetreated(butnothow theyshouldbetreated).Anexamplewouldbeusinggait datatodeterminewhethertendontransferortendon lengtheningsurgeryshouldbeperformedforaspecific childwithcerebralpalsy.Thisuseofhumanmovement dataremainsuncommonbutisbecomingmore common. 3) Prediction -Predictonanindivi dualpatientbasis whichtreatmentshouldbeperformedandhowitshould beperformed.Anexamplewouldbeusinggaitdatato determinewhethertendontransferortendonlengtheningsurgeryshouldbeperformed,whichtendonto transferorlengthen,andwheretotransferitorhow muchtolengthenit,toimprovewalkingabilityfora specificchildwithcerebralpalsy.Thisuseofhuman movementdatadoesnotyethappeninclinicalpractice.Fregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page2of11

PAGE 3

Thefocusofthispaperisonhowpersonalizedneuromusculoskeletalmodelscouldbeusedfor prediction ratherthan assessment or identification ,though identification hassignificantclinicalvalueaswell.While prediction isthemostchallenginguse,itisalsotheuse withthegreatestpotentialtoimprovefunctionaloutcomeonanindividualpatientbasis.ClinicalprocessofpersonalizedmodelingHowshouldpersonalizedneuromusculoskeletalmodels beusedtopredictfunctionaloutcomeforvarioustreatmentplansunderconsideration?Expandedfromideas presentedbyresearchersattheRizzoliOrthopedicInstituteinBologna,Italy,andDr.BartKoopmanatthe UniversityofTwenteinEnschede,theNetherlands,we proposeathree-stepprocessfortreatmentdesignusing personalizedmodels: 1) Modelpreparationsteps: Identify modeloutputstobeusedasindicatorsof clinical/functionaloutcome. Define modelcomplexityrequiredtopredictthese outputswithsufficientaccuracyfortheintended clinicalapplication. Collect pre-treatmentmovement,strength,and imagingdata(asrequired)toconstructthepersonalizedmodelandpredicttheoutputsofinterest. 2) Modelconstructionsteps: Calibrate modelgeometryandparametervaluesto whichtheoutputsofinterestaresensitiveusingpretreatmentmovement,strength,and/orimagingdata. Estimate modelparametervaluestowhichthe outputsofinterestareinsensitiveusingdata reportedintheliterature. Incorporate surgicalorrehabilitationtreatment plansunderconsiderationintothepersonalized model. 3) Modelutilizationsteps: Predict post-treatmentpatientfunctionforeach proposedtreatmentplan. Select treatmentplanandassociatedparameter valuesthatmaximizefunctionaloutcome,possibly usingnumericaloptimizationmethods. Validate personalizedmodelpredictionsusing post-treatmentfunction measuredfrompatients whosetreatmentwasnotplannedwithamodel. Implement optimalsurgicalorrehabilitationtreatmentplandesignedwiththepersonalizedmodel. Collect post-treatmentmovement,strength,and/or imagingdatafromthepatienttoassessclinical/functionaloutcome. Inthisprocess,onlythestepsrelevanttothemobilityrelatedclinicalapplicationathandneedbeperformed. Forexample,clinicalapplic ationsthatdonotrequire modelingofindividualmuscleforcesmaynotrequire anyimagingandstrengthdatafromthepatient,and thusstepsrelatedtocalibrationofpatient-specificmuscleandbonegeometrycanbeomitted.Modelparametervaluesthatrequirecalibrationtopatientdatamay necessitatecollectionofadditionalexperimentaldata solelyforcalibrationpurposes[1]. Twocriticaltaskstohighlightinthisprocessare Calibrate and Validate .Unlessthemodeliscalibratedto relevantdatacollectedfromthepatientpriortotreatment,themodelwillnotbesufficientlypersonalizedto predictthepatient spost-treatmentfunction.Similarly, unlesscalibratedmodelpredictionsarevalidatedusing post-treatmentdatacollectedfrompatientswhosetreatmentswerenotplannedwiththemodel,clinicianswill nothaveconfidenceinthemodelpredictions,andpersonalizedmodelswillneveradvancetowardwidespread clinicalutility.Validationo ftreatmentplanningusing personalizedmodelswillultimatelyrequirerandomized controlledtrials,whereoutcomesarecomparedbetween patientswhosetreatmentswereplannedwithapersonalizedmodelandthosewhosetreatmentswerenot.ClinicalapplicationsofpersonalizedmodelingDuringourtourofEuropeanresearchlabs,wesoughtto identifyclinicalapplicationswhereapersonalizedmodelingprocesssimilartotheoneoutlinedabovewas alreadybeingfollowed.Bytheendofthetour,wemade threevaluableobservationsrelatedtoclinicalapplication ofpersonalizedneuromusculoskeletalmodels.First,few labshavereachedthepointofbeingabletoapplythis processtospecificclinicalproblems.Second,someof thebestexistingclinicalapplicationsinvolvedgeneric ratherthanpersonalizedmo dels.Third,mostclinical applicationsweobservedinvolvedorthopedicsurgery, withfewapplicationsinvolvingneurorehabilitation. Belowwecommentfurtherontheseobservations. ThreelargeprojectsfundedbytheEuropeanCommission(EC)aremakingsignificantstridesindeveloping andapplyingpersonalizedneuromusculoskeletalmodels toorthopedicclinicalproblems.Thefirstisthe OsteoporoticVirtualPhysiologicalHuman (VPHOP)project [2],whichinvolvesalargeconsortiumofacademicand industrialpartnersthroughoutEuropeandiscoordinatedbyDr.MarcoVicecontiattheRizzoliOrthopedic InstituteinBologna,Italy[3-5].Asstatedontheproject website,thegoalisto develop,validateanddeploythe nextgenerationoftechnologytopredicttheabsolute riskoffractureinpatientswithlowbonemass,thereby enablingclinicianstoprov idebetterprognosesand implementmoreeffectivetreatmentstrategies. Oneof theuniqueemphasesoftheprojectisonmulti-scaleFregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page3of11

PAGE 4

modeling,withbonebeingmodeledsimultaneouslyon thecell,tissue,organ,andbodylevelstopermitclinicallyusefulpredictionsoftheriskofbonefracturein differentpatientpopulations. Inarelatedproject,Viceconti steamattheRizzoli OrthopedicInstitutehasdevelopedpersonalizedneuromusculoskelealmodelsofpediatricpatientswho receivedasurgicallimbsalvageprocedureforbonecancer[4,6].Forthisclinicalproblem,thechallengeisto determinehowthepatientshouldloadtheboneallograftduringtherehabilita tionprocesssuchthatbone loadsarehighenoughtostimulaterepairbutlow enoughtoavoidfracture.Sinceeachclinicalcaseis unique,surgicalandrehabilitationtreatmentdesigncannotbestandardized.Dr.Vicecontiandhisresearch teamareusinggaitandimagingdatatocreatepersonalizedneuromusculoskeletalmodelsthatestimatemuscle andboneloadsinthepatient sfemurduringwalking (Figure1).Theseestimatesinformtherehabilitation processandwhenthepatientshouldbeclearedforfull functionalloadingwithnorestrictions.Theprimary challengesfacedbythispersonalizedmodelingprocess arewhetherscalingofmuscleandbonegeometryfrom agenericmodelissufficientlyaccurateforthispediatric application,andalsowhethertheestimatedmuscleand boneloads(whichcurrentlycannotbevalidatedexperimentally)aresufficientlyreliable. ThesecondlargeEC-fun dedprojectiscalled NMS Physiome [7],whichisalsocoordinatedbyresearchers attheRizzoliOrthopedicInstitute.Thisprojectseeksto promoteamoreorganiccooperationinthedevelopmentofPredictive,PersonalisedandIntegrativemusculoskeletalmedicine byintegratingresearchefforts betweentheVPHOPprojectandtheCenterforPhysicsFigure1 Personalizedmodelingworkflowformassiveskeletalreconstruction,asdevelopedbytheMedicalTechnologyLaboratoryat theRizzoliOrthopedicInstituteinBologna,Italy .a)CTscanofthelowerlimbsperformedatfollow-up,withmotioncapturemarkersvisible aswell.b)Focusedviewofreconstructedfemurimmediatelyaftersurgery.c)Patient-specificmusculoskeletalmodelsuperimposedonCT images(LHPBuilder,B3C,Italy).d)Oneframeofdynamicwalkingsimulationperformedwiththepatient-specificmodel(OpenSim).Image courtesyofDr.GiordanoValente,RizzoliOrthopedicInstitute,Bologna,Italy. Fregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page4of11

PAGE 5

basedSimulationofBiologicalStructures(Simbios)at StanfordUniversityintheUnitedStates.Integrationis focusedonneuromusculoskeletalsoftwaretools(MAF, OpenSim,andFEBio)andresearchcommunitywebsites (BiomedTownandSimtk)developedbythetwoconsortia.Thegoalofintegrationistoaddressthechallenges posedbypersonalizedneuromusculoskeletalmodeling moreeffectivelyandefficiently. ThethirdlargeEC-fundedprojectisentitled ImprovingSafetyandPredictabilityofComplexMusculo-skeletalSurgeryusingaPatientSpecificNavigationSystem (TLEMsafe)[8],whichinvolvesaconsortiumofacademicandindustrialpartnersheadedbytheUniversity ofTwenteinEnschede,theNetherlands.Thestated goaloftheprojectisto createanICT-basedpatientspecificsurgicalnavigati onsystemthathelpsthesurgeonsafelyreachtheoptimalfunctionalresultforthe patientandisauserfriendlytrainingfacilityforthesurgeons. Inthisproject,theresearchersproposedtouse personalizedneuromusculoskeletalmodelsaspartofa three-steptreatmentdesignprocess.Thefirststepis creationofthepersonalizedmodelfromthepatient s movementandimagingdata.Themodeliscreated withintheframeworkoftheAnyBodymusculoskeletal modelingsoftwaredevelopedbyresearchersattheUniversityofAalborginDenmark[9](Figure2).Thesecondstepisforthesurgeontoperformvirtualsurgical treatmentsonthepersonalizedmodelandtoidentify theoptimalsurgicalplanforthepatient.Thefinalstep istotransfertheoptimizedtreatmentplanintoasurgicalnavigationsystemtobeusedduringactualsurgery. Aspartofthisproject,Dr.BartKoopmanoftheUniversityofTwenteiscurrentlyi nvestigatingtheuseofpersonalizedneuromusculoske letalmodelstoidentify optimalpatient-specifict endontransferproceduresto restorehipadductorstrengthinpatientswhowalkwith a drooping swingleghip(i.e.,Trendelenburggaitdue toPoliomyelitisortotalhiparthroplasty). Otherresearchweobservedinvolvedtheuseofgenericratherthanpersonalizedneuromusculoskeletal models.Whilepersonalizedmodelshavethegreatest potentialtoimpactclinicpractice,genericmodelscan stillprovidesignificantcli nicalbenefits.Twoclinical applicationsofgenericmod elswereparticularlywell developed.Thefirstwasthedesignofatotalankle replacementbyDr.AlbertoL eardiniandcolleaguesat theRizzoliOrthopedicInstituteinItaly[10,11].The designwasdevelopedusingasagittalplanemusculoskeletalmodeloftheanklethatincorporatedtheligaments andarticularsurfaces.Thedesignphilosophywasto maintainmedialandlateralankleligamentsinanisometricstateduringpassiveanklemotion.Theteam identifiedanovelgeometricdesigntoachievethisgoal, usingnon-anatomicallyshapedtibialandtalarcomponentswithameniscalbearinginterposedbetweenthem. Thedesignhasbeenlicensedbyanorthopedicimplant Figure2 ExamplesofmusculoskeletalmodelsdevelopedfortheTLEMsafeproject .ImagecourtesyofProf.Dr.Ir.BartKoopman,University ofTwente,Enschede,theNetherlands. Fregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page5of11

PAGE 6

company,andearlyclinicalassessmentisdemonstrating goodrestorationofanklemobilitywithlowcomplicationandrevisionrates[12,13]. Theotherexamplewasevaluationoftendontransfer surgeryformassiverotatorcufftearsbyDr.Fransvan derHelmandcolleaguesatDelftUniversityofTechnologyintheNetherlands[14,15].Theevaluationwasperformedusingahighfidelitymusculoskeletalshoulder andelbowmodelconstructedfromextensivemeasurementsperformedonasinglecadaverspecimen[16]. Themodelaccountsformore parameters(including musclesarcomerelengthmeasuredbylaserdiffraction) thananyotherupperextremitymodel.Simulationsperformedwiththegenericmodelhaveprovidedspecific recommendationsforwhichtendonstotransfer,and wheretotransferthem,toreplacethefunctionofatorn rotatorcuffwithoutsacrificingshoulderstrengthfor functionaltasks.Similartothepediatriconcologyapplication,thereliabilityofthemodel spredictedclinical outcomeisonlyasgoodasthereliabilityofthemodel s predictedmuscleforces.Dr.vanderHelmandcolleagueshaveattemptedtovalidatethemodel sprediction ofshouldermuscleandcontactforcesusingcontact forcesmeasuredbyaninstrumentedshoulderprosthesis [17].Theauthorsconcludedthat, Althoughresults indicatedareasonablecompatibilitybetweenmodeland measureddata,adjustmentswillbenecessarytoindividualizethegenericmodelwiththepatient-specific characteristics. Theprimaryrehabilitationapplicationsweobserved utilizedapersonalizedmodelingmethodcalledComplementaryLimbMotionEstimation(CLME)[18,19].The methodusesmotionmeasurementsmadeonapatient s healthylegtopredicthowthepatient simpairedleg shouldmove.Thepredictionsaremadeinrealtimeby exploitingthestrongcouplingthatexistsbetweenskeletaldegreesoffreedomduri nglocomotion.Thesecouplings(orsynergies)areiden tifiedinhealthysubjects usingstatisticaldimensionalityreductionmethods(e.g., principalcomponentanalysis)andthenappliedtothe healthylimbofapatienttopredictthedesiredmotion ofthepatient simpairedlimb.CLMEwasfirstproposed togeneratepersonalizedmotiontrajectoriesforthe pareticlimbofpatientsundergoingroboticgaittraining followingstroke[19].Thegoalwastomaintainpatient stabilitywhileminimizingunwantedinteractionforces betweenpatientandrobot.Morerecentlythesame methodhasbeenproposedtopersonalizethecontrolof anactivekneeexoprosthesistothegaitpatternsof patientswhohaveundergoneabove-kneeamputation [18].Forbothapplications,thepredictedmotionofthe impairedorprostheticlimbisusedasapersonalized referencetobetrackedbytherobotorprosthesiscontrolsystem.Whileweobservedotherpersonalized rehabilitationapplications,theydidnothaveastrong neuromusculoskeletalmodelingcomponenttothem.ClinicalgapsinpersonalizedmodelingMobility-relatedclinicalproblemsaretypicallytreated byeithersurgeryorrehabilitationandeitherdoordo notpossessasignificantneurologicalcomponent.Thus, useofpersonalizedneuromuscularmodelstoimprove treatmentofmobilityimpairmentscanbegroupedinto fourcategories:1)surgicaltreatmentwithoutasignificantneurologicalcomponent,2)surgicaltreatmentwith asignificantneurologicalcomponent,3)rehabilitation treatmentwithoutasignificantneurologicalcomponent, and4)rehabilitationtreatmentwithasignificantneurologicalcomponent.Ofthesefourcategories,thefirstis themostdevelopedandthefourththeleastdeveloped intermsofpersonalizedneuromusculoskeletalmodeling. Thissituationisnotsurprisinggiventhattechnologyis amorerecentadditiontorehabilitationtreatments(e.g., rehabilitationrobotics)tha ntosurgicaltreatmentsand thatneurologicalfactorsaremoredifficulttomodel thanaremechanicalfactors.Belowwepresentclinical gapsinpersonalizedmodelingforeachofthesefour categories. Osteoarthritisisaprevalentdisablingdiseasethatis commonlytreatedbysurgica lintervention.Thoughit clearlypossessesaneurologicalcomponent[20],that componentissecondarytomechanicalfactorsasfaras surgicaltreatmentdesignisconcerned.Useofpersonalizedneuromusculoskeleta lmodelshasbeenproposed byEuropeanlabsforpre-operativeplanningofhigh tibialosteomitiesandtotaljointreplacements[21,22]. Whilejointreplacementsur geryisgenerallyreliable, individualcasescanposespecialchallenges,especially thoseinvolvingrevisionsurgery.Incontrast,hightibial osteotomy(HTO)isachalle ngingsurgicalprocedure withhighlyvariableoutcomesbutalsohighpotential benefits,makingitanexcellenttargetforpersonalized models.Useofpersonalizedmodelstodesigncustomizedgaitmodificationsfo llowingHTOsurgerycould alsobevaluableforavoidingthelossofboneycorrectionthatoftenoccursovertime.Anteriorcruciateligamentreplacementtoavoidkneeosteoarthritisisa relatedsurgicalapplicatio nwherepersonalizedmodels couldbeofvalue[23]. Incontrast,cerebralpalsyandCharcot-Marie-Tooth diseaseareneurologicaldisordersthatarecommonly treatedbysurgery,sincenotreatmentexistsforthe underlyingneurologicalproblem.ThoughCharcotMarie-Toothdiseaseisnotwellknown,itisthemost commonlyinheritedneurologicaldisorderandlimits mobilityinapproximately1in2,500individuals[24]. Surgicaltreatmentsforbothdisorderstypicallyinvolve musclelengthening,tendontransfer,and/orosteotomy toimprovejointrangeofmotion,foot-groundcontactFregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page6of11

PAGE 7

pattern,gaitspeed,andgaitsymmetry.Forbothdisorders,patientshavevariedanduniqueclinicalpresentations,makingstereoty picaltreatmentplanning ineffective.Forthisreason,personalizedneuromusculoskeletalmodels,especiallythosethatareabletomodel theneurologicallimitations(e.g.,musclespasticity)of thepatient,couldplayaval uableroleinpredictingthe outcomeofcomplexmulti-levelsurgeriesthatareperformedonthesepatients[25-27]. Stroke,spinalcordinjury,andtraumaticbraininjury significantlyaffectmobility,possessamajorneurological component,andareoftentreatedbyrehabilitation methods.Personalizedneur omusculoskeletalmodels haveyettobeappliedtotraditionalorrobot-assisted rehabilitationtreatmentsforthesedisorders.Giventhis largegap,evenpersonalizedmodelsthatomitneural controlmodelshavethepotentialtomakeasignificant clinicalimpact.Forexample,anumberofclinicaland researchlabsinEuropeareutilizingrobot-assistedtherapyforneurorehabilitation[28,29].Manyoftheselabs areusingtheLokomatgaittrainer(HocomaAG,Volketswil,Switzerland),whoseprogrammedwalkingpattern isthatofoneofthedesigners.Personalizedmusculoskeletalmodelsthatcanpredictpatient-specificimprovementsingaitpatterncouldbeusedtocustomizerobotprescribedgaitmotions.Forindividualswhohavehada stroke,similarmodelscouldbeusedtopredicta patient-specificsequenceofgradualgaitalterationsleadingtonormalfunction,withthemodelindentifying wheretofocusrehabilitationeffortstomaximizefunctionaloutcome. Personalizedmusculoskeletalmodelsthatincludepersonalizedneuralcontrol modelswouldbeevenmore beneficialforimprovingrehabilitationofneurological disorders.Modelsthataccountforpatient-specific neuralcontrollimitationsan dneuroplasticitycouldbe usedtoidentifythemaximumexpectedimprovement andhowbesttogetthere.AssuggestedbyDr.Herman vanderKooijoftheUniversityofTwenteintheNetherlands,suchmodelscouldbeusefulforpredictinghow peopleinteractwithandada pttotheirenvironments, whichcouldimprovetheeffectivenessofrobotictherapy systems.Furthermore,suchmodelscouldbevaluablefor thedesignofneuroprosthesesthatusefunctionalelectricalstimulationtoresto relostfunction[30,31].For example,ifapersonalizedneuromusculoskeletalmodel couldpredictaminimumsetofmusclestostimulate, andhowandwhentostimulatethem,torestoreanormalgaitpattern,thenthepersonalizedprescription couldbeinvestigatedinaclinicalenvironment.Asstatedbyoneoftheresearchersonourpanel, Thereisa needfor...improvedmodelsofhumanmotorrecovery toprovideamorerationalframeworkfordesigning robotictherapycontrolstrategies. [32]. Oneoftheprimaryreasonsfortheseclinicalgapsis lackofeffectivecollaborationbetweenclinicalresearchersandpersonalizedmodelingresearchers.Anexcellent counterexampleistheRizzo liOrthopedicInstitutein Italy,wherecliniciansandengineerssharethesame officespaceandinteractduringclinicaldecisionmaking. Theseinteractionscreateanatmospherewherecliniciansroutinelyenterintothetechnicalworldandengineersroutinelyenterintotheclinicalworld.Suchan environmentofsharedintellectualinvestmentinsolving clinicalproblemsiscriticalifpersonalizedneuromusculoskeletalmodelingistomakeabroadimpactinthe clinic.TechnicalaspectsofpersonalizedmodelingSignificantresearcheffortsarecurrentlyunderwayin labsthroughoutEuropetode veloppersonalizedneuromusculoskeletalmodelingt oolsandmethods.Recalling thesectionaboveonthe ClinicalProcessofPersonalized Modeling,theprimarychallengesfacedbytheseefforts arethe Calibrate stepwithin Modelconstruction and the Validate stepwithin Modelutilization .Inthissection,wediscusscurrenttechnicalcapabilitiesofpersonalizedmodelingrelatedtomodelcalibrationand validation,followedbyadiscussionoftechnicalgaps thatneedtobefilledifpersonalizedneuromusculoskeletalmodelsaretobecomeclinicallyuseful.TechnicalcapabilitiesofpersonalizedmodelingDespitesignificantcomputationaladvancesoverthepast tenyears,modelpersonaliza tionremainsamajorchallenge,asdoestheabilitytouseapersonalizedmodelto predicttheoutcomeofaclinicalintervention.Personalized neuromusculoskeletalmodelscanbeappliedtomobilityrelatedclinicalproblemsonlytotheextenttowhichkey modelfeaturescanbecalibratedtodatacollectedfroma patient.Thus,theabilitytocalibratemodelstopatient dataisaprerequisitetoclinicaluseofpersonalizedmodels,withtheproposedclinicalapplicationdeterminingthe extentofmodelpersonalizationrequired. Sincemostneuromusculoske letalmodelsaregeneric, beingconstructedfromdetailedanatomicmeasurements performedoncadaverspecimens[16,33],amodelpersonalization(orcalibration)processisneeded.Expanding oninformationprovidedbyDr.BartKoopmanofthe UniversityofTwenteintheNetherlands,wepropose fourmodelcalibrationstepsthatshouldbeperformed inwholeorinparttotransformagenericmodelintoa personalizedmodel: 1) Geometriccalibration -Useofimagingdata(e.g., MR,CT,x-ray)tocalibratebonegeometry,musclelines ofaction,andmusclemomentarmsinamusculoskeletalmodel. 2) Kinematiccalibration -Useofmotiondata(e.g., marker-based,inertialsenso rs,fluoroscopy)tocalibrateFregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page7of11

PAGE 8

constraint-basedjointpositionsandorientationsinthe bodysegmentsofaskeletalmodel. 3) Kineticcalibration -Useofloaddata(e.g.,ground reactionforceandmoment,footcontactpressure, dynamometer)tocalibratebodysegmentmassandinertia,footstiffness,musclestrength,andothermuscletendonpropertiesinaneuromusculoskeletalmodel. 4) Neurologiccalibration -Useofmotion,load,and muscleactivitydata(i.e.,muscleEMG)tocalibratefeedforward,intrinsicfeedback,reflexivefeedback,and/or synergypropertiesoftheneuralcontrolsystemina neuromusculoskeletalmodel. Thechallengeishowtoconstructapersonalized modelthatisconsistentwithallavailabledatafrom thesedifferentmodalities[34]. Currentmethodsforgeometriccalibrationinvolve uniformscaling,non-uniformscaling,deformation,or directcreationofbonemodelsandmusclelinesof actionfrompatientMRorCTdata.Uniformscaling basedonexternalmeasurementsisinaccuratewhencalculatingmusclemomentarms,muscle-tendonlengths, muscleforces,andjointcontactforces[35],especially whenscalingagenericmodelofanadulttoapediatric patient[36].Non-uniformscalingisonlyslightlybetter atproducingaccuratemusclemomentarmsandmuscle-tendonlengths[27].Creationofpatient-specificgeometrydirectlyfromthepatient simagingdataremains thegoldstandard[26],buttheprocessishighlytime consumingandsomewhatsubj ective,dependingonthe imagingmodalityandtheanatomicstructuresbeing modeled(e.g.,boneedgesareoftenpoorlydefinedin MRdata). Kinematiccalibrationinvolvesdeterminingfixedjoint positionsandorientationsinthebodysegmentsofa skeletalmodelwithapre-definedkinematicstructure. Thecalibrationprocessisusuallyperformedusingsurfacemarkerdata,withjointanglesinthemodelbeing calculatedasabyproduct.Europeanlabshaveusedoptimizationmethods[1]andextendedKalmanfiltermethods[37-39]toperformkinematiccalibration.Filterbasedmethodshavetheadvantageofbeingcomputationallyfasterandlesscomplexthanmostoptimization methods,buttheyrequireagreateramountofalgorithm tuningtoachievesatisfactoryperformance.Despite theseadvances,neitherapproachhasyettobegeneralizedandincorporatedintocommercial-grademusculoskeletalmodelingsoftwaresuchastheAnyBody program. Kineticcalibrationtypicallyinvolvescalibrationofsegmentmasspropertiesormuscleforce-generatingproperties.Thoughsegmentmasspropertiescanbe calibratedtoforceplateandmotiondata[40],theyare oftentakenfromregressionequationsdevelopedfrom measurementsperformedoncadavers[41].Similarly, thoughmusclemodelparametervalues(e.g.,muscle strength)canbecalibratedtoisometricand/orisokinetic dynamometerdata[42],thiscalibrationstepisusually omittedduetotheextraeffortitrequires.ArecentEuropeanstudyindicatedthatsubj ect-specificmuscle-tendonparametervaluescalibratedtodynamometerdata areappropriateforuseinmusculoskeletalmodelsused toanalyzegait[43].Sincemanymovementimpairments involveundesirablefoot-groundcontactpatterns,kinetic calibrationofpatient-specificfoot-groundcontactmodelswillbeessentialinthefutureforpredictingchanges ingaitfunctionduetovariousproposedtreatments,yet kineticcalibrationmethodsforsuchmodelsdonotyet exist. Apromisingdevelopmenttoaddressgeometric,kinematic,andkineticcalibrationsimultaneouslyisresearch beingperformedbyDr.WafiSkalliatArtsetMtiers ParisTechinParis,FranceusingtheEOSbi-planex-ray system(EOSImagingSA,Paris,France).WiththeEOS system,asubjectstandsupr ightwhilealow-dosex-ray systemscanstheentirebodyfromheadtotoecollecting onecontinuousdistortion-freeimageineachoftwo orthogonalplanes.Theresultingbi-planeimagesare thenprocessedandmorphedusingtemplateanatomy forpersonalizationandvisualization.Geometriccalibrationofboneandmusclegeometryviadeformationof templateboneandmusclemodelscanbeperformed rapidlyandaccuratelyrelativetogeometryconstructed directlyfromCTdata[44-47].Kinematiccalibration couldtheoreticallybeaidedbyperformingscansofthe relevantportionofthebodyintwoormoreposes(e.g., thelowerextremitiesindifferentsquattingpositions) [48],especiallyifsurfacemarkerstobeusedinadditionalmovementexperiment sarealsovisible.Kinetic calibrationofsegmentmasspropertiesandmuscle strengthparameters(basedonmusclecrosssectional areas)canalsobeperformedfromtheimages[49,50]. Thus,withimprovementsinautomationandrefinement ofexistingalgorithms,theEOSsystemhasthepotential toimprovemusculoskeletalmodelpersonalization significantly. Theremainingarea,neurologiccalibration,hasseen theleastprogressduetothesignificantchallenges involvedinunderstandinghowthehumanneuralcontrolsystemfunctions.Thiscalibrationstepcanbepursuedusingarangeofapproaches,fromaphysiological approachthatseekstomodelthedetailedanatomyand physiologyinvolvedinneuralcontrol,toanemergent approachthatseekstomodeltheneuralcontrolcomputationsimplementedbytheanatomybutwithoutmodelinganatomicdetail.Anexampleofaphysiological approachisdetailedmodelingoffeedforward,intrinsic (i.e.,muscle)feedback,andreflexive(i.e.,visual,proprioceptive,andvesitibular)f eedbackmechanismsutilizedFregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page8of11

PAGE 9

bytheneuralcontrolsystem[51-54].Todate,suchhigh fidelityneuralcontrolmodelshavebeenappliedtoposturalcontrolratherthanmovementtasks,andmethods forpersonalizingtheparametervaluesinthesemodels arenotyetwelldeveloped.Attheotherextreme,an exampleofanemergentapproachismusclesynergy analysis,whereEMGsignalsfromalargenumberof muscles(e.g.,16)aredecomposedintoasmallernumberofbasisactivationsignals(e.g.,5)forallmuscles plusauniquesetofweights(oftentermed modules ) foreachmusclethatscaletheactivationsignals[55,56]. Synergyanalysisisusedfordimensionalityreduction(e. g.,5basissignalsareusedtoreconstruct16EMGsignals)andcanidentifyneuralcontrollimitationsin patientsfollowingstroke[57].Incorporationofthese limitationsintopersonalizedneuromusculoskeletalmodelscouldfacilitatepredictionofbestpossiblefunctional outcome.Betweenthesetwoextremesisaphysiological approachthathassuccessfullyexplainedmotorlearning usingsimplifiedfeedforwa rdandfeedbackmodels[58]. Theapproachusesav-shapedlearningfunctionto modelthechangeinmusclefeedforwardcommands generatedinresponsetokinematicerrorsexperienced duringthepreviousmovementtrial.Computersimulationofasequenceofarmmovementtrialsperformedin differentforcefieldsrevealedthatthemethodcansuccessfullyreproduceexperimentallyobservedtrial-to-trial changesinmuscleactivations(tocontrolforce)andcocontraction(tocontrolimpedence). Validationofclinicalpredi ctionsistheothermajor challengefacedbypersonalizedmodels.Thischallenge canbesurmountedwhenclinicaloutcomevariablesare externalquantitiesthatcanbeeasilymeasured(e.g.,gait speed,gaitsymmetry).Frequently,however,established clinicaldatabasesusecoarseordinalscalestorank movementability,andmappi ngthesescorestoneuromechanicalmodelsisdifficul t.Inaddition,significant challengesremainwhentheoutcomevariablesareeither internaltothebody(e.g.,m uscleforces,jointcontact forces,bonestrains)ordependentonquantitiesthatare internaltothebody.Sinces uchquantitiescannotbe measureddirectlybynon-invasivemeans,alternate methodsareneededforperso nalizedmodelvalidation. Forexample,predictedmuscleforceshavebeenevaluatedindirectlyusinginvivojointcontactforcemeasurements[17]ornovelmeasurements(e.g.,nearinfrared spectroscopy)thatarelikelytobehighlycorrelatedwith invivomuscleforce[59].TechnicalgapsinpersonalizedmodelingAssuggestedbythisreviewofcurrenttechnicalcapabilitiesinEurope,atleastfourcriticaltechnicalgapscurrentlyexistthatlimitthepot entialclinicalapplicability ofpersonalizedneuromusculoskeletalmodels: 1) Howcanwemakethepersonalizedmodelcalibrationandpredictionprocessfastandeasy? Thoughseveralexcellentmusculoskeletalmodeling programsexist,noneofthemcontainfunctionality thatautomatesthemodelcalibrationprocessand simplifiesthemodelpredictionprocess.Personalized modelcalibrationandpredi ctioncurrentlyrequire significantexpertiseandprogrammingabilitypossessedbyonlyasmallnumberofresearchersina limitednumberofresearchlabs.Makingthesecapabilitiesavailabletothelargerneuromusculoskeletal modelingcommunityviafast,automatedalgorithms willbeessentialforthegrowthofpersonalizedmodelingefforts.Ultimately,personalizedmodelingwill beadoptedforroutineclinicaluseonlywhenitis extremelyeasytouse. 2) Howcanwecalibrate unobservable parametersto whichmodelpredictionsaresensitive? Forsomeclinicalproblems,personalizedmodelpredictionsoffunctionaloutcomewillbesensitiveto modelparametervaluesthatcannotbecalibratedto availabledata.Thefirststepinaddressingthisproblemisidentifyingwhenitoccurs,whichrequires performingsensitivityanalysesthatinsomecases willbelimitedbyexistingcomputationalcapabilities. Thenextstepisdevelopmentofnewexperimental methodsorhardwarethatprovidesufficientlyrich informationtocalibratetheparametervaluesneeded todevelopthepredictions. 3) Howcanwecreatepersonalizedneuralcontrol models? Fewneuromusculoskeletalmodelspublishedtodate accountforanylevelofpersonalizedneuralcontrol modeling.Suchmodelingwouldideallyaccountfor limitationsinapatient sneuralcontrolcapabilities aswellastheextentofpossibleplasticity.Emergent approachesformodelingneuralcontrolcouldbe incorporatedintopersonalizedmusculoskeletalmodelsasastartingpoint,whilephysiologicalmodels couldberefinedtothepointwhereessentialmodel parametersbecomewelldefinedandmethodsfor calibratingthemaredeveloped.Theabilitytoincorporatecomplexpersonalizedneuralcontrolmodels intopersonalizedmusculoskeletalmodelswould greatlyexpandmodelapplicabilitytoclinicalsituations,especiallythoseinvolvingneurorehabilitation. 4) Howcanwevalidatemodel-basedpredictions,Fregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page9of11

PAGE 10

especiallyforinternalquant itiessuchasmuscle,joint, andboneloads? Validationofinternalquant itiesthatinfluencetreatmentdesignremainsamajorchallenge.WhileresearcherscontinuetorefineoptimizationandEMG-driven methodsforpredictingmuscleforcesandrelatedjoint andboneloads,theabilitytovalidatethesepredictions haslaggedbehind.Directmeasurementofinternal quantitiesunderspecialconditions(e.g.,instrumented implants),andtheopportunitytotestmodel-basedpredictionsagainsttheseinternalmeasurements,providesa valuableavenueformodelvalidationefforts[60].Identificationofnovelapproachesthatutilizeonlyexisting datacollectioncapabilities,aswellasdevelopmentof newexperimentaltechniques ,willbeessentialifcliniciansaretogainconfidenceintreatmentplansdesigned withpersonalizedneuromusculoskeletalmodels.ConclusionsNeuromusculoskeletalmodelinghasyettomakeasignificantdifferenceinroutineclinicalpractice.Forthis situationtochange,thekeygapsidentifiedaboveneed tobeaddressedbymodelingresearchersinclosecollaborationwithclinicalinvestigators.Whilethebiggest clinicalgapforpersonalizedneuromusculoskeletalmodelingisinneurorehabilitation,thegapforothermobility-relatedclinicalprobl emsisalmostaslarge.The biggesttechnicalgapisinpersonalizedneuralcontrol andrecoverymodels,thoughissueslikeautomationof themodelpersonalizationprocessanddevelopmentof personalizedfoot-groundcon tactmodelsarecriticalas wellforadvancement.Forclinicalproblemsthatinvolve highlyuniquepatientcharacteristics,stereotypicaltreatmentdesignislikelytoyield variablefunctionaloutcomes.Thesetypesofclinicalproblemsarewhere personalizedneuromusculoskeletalmodelshavethe greatestpotentialtocreateapositiveparadigmshiftin thetreatmentdesignprocess.Acknowledgements TheauthorswishtothankDr.TedConwayoftheNationalScience FoundationforfundingthisreviewofEuropeanresearchandWorld TechnologyEvaluationCenterformanagingthelogisticsofthestudytour. Authordetails1DepartmentsofMechanical&AerospaceEngineering,Biomedical Engineering,andOrthopaedics&Rehabilitation,UniversityofFlorida,231 MAE-ABuilding,P.O.Box116250,Gainesville,FL32611-6250,USA.2DepartmentofPhysicalMedicine&Rehabilitation,UniversityofPittsburgh, Pittsburgh,PA,USA.3DepartmentsofMechanical&AerospaceEngineering andBiomedicalEngineering,UniversityofCalifornia,Irvine,CA,USA. Authors contributions BForganizedanddraftedthemanuscriptbasedoninformationgathered duringtheEuropeantour.MLprovidedclinicalperspectiveoninformation gathered.DRorganizedthetourandtheselectionofEuropeansites.All authorsmadesignificantcontributionstogathering,criticallyevaluating, organizing,andrevisingthecontentofthemanuscript.Allauthorsreadand approvedthefinalmanuscript. Competinginterests Theauthorsdeclarethattheyhavenocompetinginterests. Received:29September2011Accepted:30March2012 Published:30March2012 References1.AndersenMS,DamsgaardM,MacWilliamsB,RasmussenJ: A computationallyefficientoptimisation-basedmethodforparameter identificationofkinematicallydeterminateandover-determinate biomechanicalsystems. ComputMethodsBiomechBiomedEng 2010, 13 :171-183. 2. OsteoporoticVirtualPhysiologicalHumanProject. [http://www.vphop.eu/ ]. 3.VicecontiM,KohlP: Thevirtualphysiologicalhuman:computer simulationforintegrativebiomedicineI. PhilTransAMathPhysEngSci 2010, 368 :2591-2594. 4.VicecontiM,TaddeiF,VanSintJanS,LeardiniA,CristofoliniL,SteaS, BaruffaldiF,BaleaniM: Multiscalemodellingoftheskeletonforthe predictionoftheriskoffracture. ClinBiomech 2008, 23 :845-852. 5.VicecontiM,ClapworthyG,VanSintJanS: TheVirtualPhysiological Human-aEuropeaninitiativeforinsilicohumanmodelling. JPhysiolSci 2008, 58 :441-446. 6.TaddeiF,MartelliS,ValenteG,LeardiniA,BenedettiMG,ManfriniM, VicecontiM: Femoralloadsduringgaitinapatientwithmassiveskeletal reconstruction. ClinBiomech 2011,(conditionallyaccepted). 7. NMSPhysiomeProject. [http://www.nmsphysiome.eu/]. 8. ImprovingSafetyandPredictabilityofComplexMusculo-skeletalSurgery usingaPatient-SpecificNavigationSystem(TLEMsafeProject). [http:// www.tlemsafe.eu/]. 9.DamsgaardM,RasmussenJ,ChristensenST,SurmaE,deZeeJ: Analysisof musculoskeletalsystemsintheAnyBodymodelingsystem. SimulModel PracTheory 2006, 14 :1100-1111. 10.LeardiniA,CataniF,GianniniS,O ConnorJJ: Computer-assisteddesignof thesagittalshapesofaligament-compatibletotalanklereplacement. MedBiolEngComput 2001, 39 :168-175. 11.LeardiniA,O ConnorJJ,CataniF,GianniniS: Mobilityofthehumanankle andthedesignoftotalanklereplacement. ClinOrthopRelatRes 2004, 424 :39-6. 12.IngrossoS,BenedettiMG,LeardiniA,CasanelliS,SforzaT,GianniniS: Gait analysisinpatientsoperatedwithanoveltotalankleprosthesis. Gait Posture 2009, 30 :132-137. 13.GianniniS,RomagnoliM,O ConnorJJ,MalerbaF,LeardiniA: Totalankle replacementcompatiblewithligamentfunctionproducesmobility, goodclinicalscores,andlowcomplicationrates:anearlyclinical assessment. ClinOrthopRelatRes 2010, 468 :2746-2753. 14.MagermansDJ,ChadwickEK,VeegerHE,vanderHelmFC,RozingPM: Biomechanicalanalysisoftendontransfersformassiverotatorcufftears. ClinBiomech 2004, 19 :350-357. 15.MagermansDJ,ChadwickEK,VeegerHE,RozingPM,vanderHelmFC: Effectivenessoftendontransfersformassiverotatorcufftears:a simulationstudy. ClinBiomech 2004, 19 :116-122. 16. KleinBretelerMD,SpoorCW,VanderHelmFC: Measuringmuscleand jointgeometryparametersofashoulderformodelingpurposes. J Biomech 1999, 32 :1191-1197. 17.NikooyanAA,VeegerHE,WesterhoffP,GraichenF,BergmannG,vander HelmFC: ValidationoftheDelftShoulderandElbowModelusinginvivoglenohumeraljointcontactforces. JBiomech 2010, 43 :3007-3014. 18.ValleryH,BurgkartR,HartmannC,MitternachtJ,RienerR,BussM: Complementarylimbmotionestimationforthecontrolofactiveknee prostheses. BiomedTech(Berl) 2011, 56 :45-51. 19.ValleryH,vanAsseldonkEH,BussM,vanderKooijH: Referencetrajectory generationforrehabilitationrobots:complementarylimbmotion estimation. IEEETransNeuralSystRehabilEng 2009, 17 :23-30. 20.Palmieri-SmithRM,ThomasAC: Aneuromuscularmechanismof posttraumaticosteoarthritisassociatedwithACLinjury. ExercSportSci Rev 2009, 37 :147-153.Fregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page10of11

PAGE 11

21.HellerMO,MatziolisG,KonigC,TaylorWR,HinterwimmerS,GraichenH, HegeHC,BergmannG,PerkaC,DudaGN: Musculoskeletalbiomechanics ofthekneejoint.Principlesofpreoperativeplanningforosteotomyand jointreplacement. Orthopade 2007, 36 :628-634. 22.HellerMO,SchroderJH,MatziolisG,SharenkovA,TaylorWR,PerkaC, DudaGN: Musculoskeletalloadanalysis.Abiomechanicalexplanationfor clinicalresults andmore? Orthopade 2007, 36 :188-194. 23.FreyM,RienerR,MichasC,RegenfelderF,BurgkartR: Elasticpropertiesof anintactandACL-rupturedkneejoint:measurement,mathematical modelling,andhapticrendering. JBiomech 2006, 39 :1371-1382. 24.ReillyMM,MurphySM,LauraM: Charcot-Marie-Toothdisease. JPeripher NervSyst 2011, 16 :1-14. 25.vanderKrogtMM,DoorenboschCA,HarlaarJ: Validationofhamstrings musculoskeletalmodelingbycalculatingpeakhamstringslengthat differenthipangles. JBiomech 2008, 41 :1022-1028. 26.ScheysL,DesloovereK,SuetensP,JonkersI: Levelofsubject-specific detailinmusculoskeletalmodelsaffectshipmomentarmlength calculationduringgaitinpediatricsubjectswithincreasedfemoral anteversion. JBiomech 2011, 44 :1346-1353. 27.ScheysL,SpaepenA,SuetensP,JonkersI: Calculatedmoment-armand muscle-tendonlengthsduringgaitdiffersubstantiallyusingMRbased versusrescaledgenericlower-limbmusculoskeletalmodels. GaitPosture 2008, 28 :640-648. 28.VenemanJF,KruidhofR,HekmanEE,EkkelenkampR,VanAsseldonkEH,van derKooijH: DesignandevaluationoftheLOPESexoskeletonrobotfor interactivegaitrehabilitation. IEEETransNeuralSystRehabilEng 2007, 15 :379-386. 29.Duschau-WickeA,CaprezA,RienerR: Patient-cooperativecontrol increasesactiveparticipationofindividualswithSCIduringrobot-aided gaittraining. JNeuroengRehabil 2010, 7 :43. 30.RienerR: Model-baseddevelopmentofneuroprosthesisforparaplegic patients. PhilosTransRSocLondBBiolSci 1999, 354 :877-894. 31.KirschRF,AcostaAM,vanderHelmFC,RotteveelRJ,CashLA: Model-based developmentofneuroprosthesesforrestoringproximalarmfunction. J RehabilResDev 2001, 38 :619-626. 32.Marchal-CrespoL,ReinkensmeyerDJ: Reviewofcontrolstrategiesfor roboticmovementtrainingafterneurologicinjury. JNeuroengRehabil 2009, 6 :20. 33.KleinHorsmanMD,KoopmanHF,vanderHelmFC,ProseLP,VeegerHE: Morphologicalmuscleandjointparametersformusculoskeletal modellingofthelowerextremity. ClinBiomech 2007, 22 :239-247. 34.LeardiniA,BelvedereC,AstolfiL,FantozziS,VicecontiM,TaddeiF,EnsiniA, BenedettiMG,CataniF: Anewsoftwaretoolfor3Dmotionanalysesof themusculo-skeletalsystem. ClinBiomech 2006, 21 :870-879. 35.LenaertsG,BartelsW,GelaudeF,MulierM,SpaepenA,VanderPerreG, JonkersI: Subject-specifichipgeometryandhipjointcentrelocation affectscalculatedcontactforcesatthehipduringgait. JBiomech 2009, 42 :1246-1251. 36.ScheysL,VanCampenhoutA,SpaepenA,SuetensP,JonkersI: PersonalizedMR-basedmusculoskeletalmodelscomparedtorescaled genericmodelsinthepresenceofincreasedfemoralanteversion:effect onhipmomentarmlengths. GaitPosture 2008, 28 :358-365. 37.CerveriP,PedottiA,FerrignoG: Kinematicalmodelstoreducetheeffect ofskinartifactsonmarker-basedhumanmotionestimation. JBiomech 2005, 38 :2228-2236. 38.DeGrooteF,DeLaetT,JonkersI,DeSchutterJ: Kalmansmoothing improvestheestimationofjointkinematicsandkineticsinmarkerbasedhumangaitanalysis. JBiomech 2008, 41 :3390-3398. 39.HalvorsenK,JohnstonC,BackW,StokesV,LanshammarH: Trackingthe motionofhiddensegmentsusingkinematicconstraintsandKalman filtering. JBiomechEng 2008, 130 :011012. 40.LenziD,CappelloA,ChiariL: Influenceofbodysegmentparametersand modelingassumptionsontheestimateofcenterofmasstrajectory. J Biomech 2003, 36 :1335-1341. 41.RaoG,AmarantiniD,BertonE,FavierD: Influenceofbodysegments parametersestimationmodelsoninversedynamicssolutionsduring gait. JBiomech 2006, 39 :1531-1536. 42.DoorenboschCA,JoostenA,HarlaarJ: CalibrationofEMGtoforcefor kneemusclesisapplicablewithsubmaximalvoluntarycontractions. J ElectromyogrKinesiol 2005, 15 :429-435. 43.DeGrooteF,VanCampenA,JonkersI,DeSchutterJ: Sensitivityof dynamicsimulationsofgaitanddynamometerexperimentstohill musclemodelparametersofkneeflexorsandextensors. JBiomech 2010, 43 :1876-1883. 44.JolivetE,DaguetE,PomeroV,BonneauD,LaredoJD,SkalliW: Volumic patient-specificreconstructionofmuscularsystembasedonareduced datasetofmedicalimages. ComputMethodsBiomechBiomedEng 2008, 11 :281-290. 45.HumbertL,DeGuiseJA,AubertB,GodboutB,SkalliW: 3Dreconstruction ofthespinefrombiplanarX-raysusingparametricmodelsbasedon transversalandlongitudinalinferences. MedEngPhys 2009, 31 :681-687. 46.MittonD,DeschenesS,LaporteS,GodboutB,BertrandS,deGuiseJA, SkalliW: 3Dreconstructionofthepelvisfrombi-planarradiography. ComputMethodsBiomechBiomedEng 2006, 9 :1-5. 47.ChaibiY,CressonT,AubertB,HausselleJ,NeyretP,HaugerO,deGuiseJA, SkalliW: Fast3Dreconstructionofthelowerlimbusingaparametric modelandstatisticalinferencesandclinicalmeasurementscalculation frombiplanarX-rays. ComputMethodsBiomechBiomedEng 2011. 48.BergaminiE,PilletH,HausselleJ,ThoreuxP,GuerardS,CamomillaV, CappozzoA,SkalliW: Tibio-femoraljointconstraintsforbonepose estimationduringmovementusingmulti-bodyoptimization. GaitPosture 2011, 33 :706-711. 49.DumasR,AissaouiR,MittonD,SkalliW,deGuiseJA: Personalizedbody segmentparametersfrombiplanarlow-doseradiography. IEEETrans BiomedEng 2005, 52 :1756-1763. 50.SandozB,LaporteS,SkalliW,MittonD: Subject-specificbodysegment parameters estimationusingbiplanarX-rays:afeasibilitystudy. Comput MethodsBiomechBiomedEng 2010, 13 :649-654. 51.vanderKooijH,PeterkaRJ: Non-linearstimulus-responsebehaviorofthe humanstancecontrolsystemispredictedbyoptimizationofasystem withsensoryandmotornoise. JComputNeurosci 2011. 52.vanderKooijH,JacobsR,KoopmanB,vanderHelmF: Anadaptivemodel ofsensoryintegrationinadynamicenvironmentappliedtohuman stancecontrol. BiolCybern 2001, 84 :103-115. 53.SchoutenAC,deVlugtE,vanHiltenJJ,vanderHelmFC: Quantifying proprioceptivereflexesduringpositioncontrolofthehumanarm. IEEE TransBiomedEng 2008, 55 :311-321. 54.StienenAH,SchoutenAC,SchuurmansJ,vanderHelmFC: Analysisof reflexmodulationwithabiologicallyrealisticneuralnetwork. JComput Neurosci 2007, 23 :333-348. 55.IvanenkoYP,PoppeleRE,LacquanitiF: Fivebasicmuscleactivation patternsaccountformuscleactivityduringhumanlocomotion. JPhysiol 2004, 556 :267-282. 56.IvanenkoYP,PoppeleRE,LacquanitiF: Motorcontrolprogramsand walking. Neuroscientist 2006, 12 :339-348. 57.GizziL,NielsenJF,FeliciF,IvanenkoYP,FarinaD: Impulsesofactivation butnotmotormodulesarepreservedinthelocomotionofsubacute strokepatients. JNeurophysiol 2011. 58.FranklinDW,BurdetE,TeeKP,OsuR,ChewCM,MilnerTE,KawatoM: CNS learnsstable,accurate,andefficientmovementsusingasimple algorithm. JNeurosci 2008, 28 :11165-11173. 59.PraagmanM,ChadwickEK,vanderHelmFC,VeegerHE: Therelationship betweentwodifferentmechanicalcostfunctionsandmuscleoxygen consumption. JBiomech 2006, 39 :758-765. 60.FreglyBJ,BesierTF,LloydDG,DelpSL,BanksSA,PandyMG,D LimaDD: GrandChallengeCompetitiontoPredictInVivoKneeLoads. JOrthop Res 2012, 30 :503-513.doi:10.1186/1743-0003-9-18 Citethisarticleas: Fregly etal .: Personalizedneuromusculoskeletal modelingtoimprovetreatmentofmobilityimpairments:aperspective fromEuropeanresearchsites. JournalofNeuroEngineeringand Rehabilitation 2012 9 :18.Fregly etal JournalofNeuroEngineeringandRehabilitation 2012, 9 :18 http://www.jneuroengrehab.com/content/9/1/18 Page11of11


xml version 1.0 encoding utf-8 standalone no
mets ID sort-mets_mets OBJID sword-mets LABEL DSpace SWORD Item PROFILE METS SIP Profile xmlns http:www.loc.govMETS
xmlns:xlink http:www.w3.org1999xlink xmlns:xsi http:www.w3.org2001XMLSchema-instance
xsi:schemaLocation http:www.loc.govstandardsmetsmets.xsd
metsHdr CREATEDATE 2012-05-02T20:09:28
agent ROLE CUSTODIAN TYPE ORGANIZATION
name BioMed Central
dmdSec sword-mets-dmd-1 GROUPID sword-mets-dmd-1_group-1
mdWrap SWAP Metadata MDTYPE OTHER OTHERMDTYPE EPDCX MIMETYPE textxml
xmlData
epdcx:descriptionSet xmlns:epdcx http:purl.orgeprintepdcx2006-11-16 xmlns:MIOJAVI
http:purl.orgeprintepdcxxsd2006-11-16epdcx.xsd
epdcx:description epdcx:resourceId sword-mets-epdcx-1
epdcx:statement epdcx:propertyURI http:purl.orgdcelements1.1type epdcx:valueURI http:purl.orgeprintentityTypeScholarlyWork
http:purl.orgdcelements1.1title
epdcx:valueString Personalized neuromusculoskeletal modeling to improve treatment of mobility impairments: a perspective from European research sites
http:purl.orgdctermsabstract
Abstract
Mobility impairments due to injury or disease have a significant impact on quality of life. Consequently, development of effective treatments to restore or replace lost function is an important societal challenge. In current clinical practice, a treatment plan is often selected from a standard menu of options rather than customized to the unique characteristics of the patient. Furthermore, the treatment selection process is normally based on subjective clinical experience rather than objective prediction of post-treatment function. The net result is treatment methods that are less effective than desired at restoring lost function. This paper discusses the possible use of personalized neuromusculoskeletal computer models to improve customization, objectivity, and ultimately effectiveness of treatments for mobility impairments. The discussion is based on information gathered from academic and industrial research sites throughout Europe, and both clinical and technical aspects of personalized neuromusculoskeletal modeling are explored. On the clinical front, we discuss the purpose and process of personalized neuromusculoskeletal modeling, the application of personalized models to clinical problems, and gaps in clinical application. On the technical front, we discuss current capabilities of personalized neuromusculoskeletal models along with technical gaps that limit future clinical application. We conclude by summarizing recommendations for future research efforts that would allow personalized neuromusculoskeletal models to make the greatest impact possible on treatment design for mobility impairments.
http:purl.orgdcelements1.1creator
Fregly, Benjamin J
Boninger, Michael L
Reinkensmeyer, David J
http:purl.orgeprinttermsisExpressedAs epdcx:valueRef sword-mets-expr-1
http:purl.orgeprintentityTypeExpression
http:purl.orgdcelements1.1language epdcx:vesURI http:purl.orgdctermsRFC3066
en
http:purl.orgeprinttermsType
http:purl.orgeprinttypeJournalArticle
http:purl.orgdctermsavailable
epdcx:sesURI http:purl.orgdctermsW3CDTF 2012-03-30
http:purl.orgdcelements1.1publisher
BioMed Central Ltd
http:purl.orgeprinttermsstatus http:purl.orgeprinttermsStatus
http:purl.orgeprintstatusPeerReviewed
http:purl.orgeprinttermscopyrightHolder
Fregly et al.; licensee BioMed Central Ltd.
http:purl.orgdctermslicense
http://creativecommons.org/licenses/by/2.0
http:purl.orgdctermsaccessRights http:purl.orgeprinttermsAccessRights
http:purl.orgeprintaccessRightsOpenAccess
http:purl.orgeprinttermsbibliographicCitation
Journal of NeuroEngineering and Rehabilitation. 2012 Mar 30;9(1):18
http:purl.orgdcelements1.1identifier
http:purl.orgdctermsURI http://dx.doi.org/10.1186/1743-0003-9-18
fileSec
fileGrp sword-mets-fgrp-1 USE CONTENT
file sword-mets-fgid-0 sword-mets-file-1
FLocat LOCTYPE URL xlink:href 1743-0003-9-18.xml
sword-mets-fgid-1 sword-mets-file-2 applicationpdf
1743-0003-9-18.pdf
structMap sword-mets-struct-1 structure LOGICAL
div sword-mets-div-1 DMDID Object
sword-mets-div-2 File
fptr FILEID
sword-mets-div-3


!DOCTYPE art SYSTEM 'http:www.biomedcentral.comxmlarticle.dtd'
ui 1743-0003-9-18
ji 1743-0003
fm
dochead Review
bibl
title p Personalized neuromusculoskeletal modeling to improve treatment of mobility impairments: a perspective from European research sites
aug
au id A1 ca yes snm Freglymi Jfnm Benjamininsr iid I1 email fregly@ufl.edu
A2 BoningerLMichaelI2 boninger@upmc.edu
A3 ReinkensmeyerJDavidI3 dreinken@uci.edu
insg
ins Departments of Mechanical & Aerospace Engineering, Biomedical Engineering, and Orthopaedics & Rehabilitation, University of Florida, 231 MAE-A Building, P.O. Box 116250, Gainesville, FL 32611-6250, USA
Department of Physical Medicine & Rehabilitation, University of Pittsburgh, Pittsburgh, PA, USA
Departments of Mechanical & Aerospace Engineering and Biomedical Engineering, University of California, Irvine, CA, USA
source Journal of NeuroEngineering and Rehabilitation
issn 1743-0003
pubdate 2012
volume 9
issue 1
fpage 18
url http://www.jneuroengrehab.com/content/9/1/18
xrefbib pubidlist pubid idtype doi 10.1186/1743-0003-9-18pmpid 22463378
history rec date day 29month 9year 2011acc 3032012pub 3032012cpyrt 2012collab Fregly et al; licensee BioMed Central Ltd.note This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
kwdg kwd Musculoskeletal modelNeural control modelOrthopedic surgeryNeurorehabilitationBiomechanics
abs
sec st Abstract
Mobility impairments due to injury or disease have a significant impact on quality of life. Consequently, development of effective treatments to restore or replace lost function is an important societal challenge. In current clinical practice, a treatment plan is often selected from a standard menu of options rather than customized to the unique characteristics of the patient. Furthermore, the treatment selection process is normally based on subjective clinical experience rather than objective prediction of post-treatment function. The net result is treatment methods that are less effective than desired at restoring lost function. This paper discusses the possible use of personalized neuromusculoskeletal computer models to improve customization, objectivity, and ultimately effectiveness of treatments for mobility impairments. The discussion is based on information gathered from academic and industrial research sites throughout Europe, and both clinical and technical aspects of personalized neuromusculoskeletal modeling are explored. On the clinical front, we discuss the purpose and process of personalized neuromusculoskeletal modeling, the application of personalized models to clinical problems, and gaps in clinical application. On the technical front, we discuss current capabilities of personalized neuromusculoskeletal models along with technical gaps that limit future clinical application. We conclude by summarizing recommendations for future research efforts that would allow personalized neuromusculoskeletal models to make the greatest impact possible on treatment design for mobility impairments.
meta classifications classification mobility_technology subtype theme_series_title type BMC Major trends in mobility technologytheme_series_editor Brian B Caulfield and Silvestro Micerabdy
Introduction
Mobility involves walking, stair climbing, posture, balance, manipulation, transfers, and other locomotion tasks and is therefore central to qualify of life. When an individual incurs a mobility impairment, quality of life is diminished in proportion to the extent of the impairment. For example, mild knee osteoarthritis can limit participation in desired recreational or athletic activities without significantly affecting normal daily activities and productivity. In contrast, a stroke can make it nearly impossible to walk or manipulate objects, significantly diminishing an individual's ability to be self sufficient and function in society. Spinal cord injury can leave a person with normal upper extremity function but no remaining lower extremity function, significantly impacting only certain aspects of mobility.
Treatments for different mobility impairments are typically stereotypical, with a standard menu of treatment options existing for any particular mobility impairment. For example, severe medial compartment knee osteoarthritis may be treated surgically using high tibial osteotomy, unicondylar knee replacement, or total knee replacement. Once a patient seeks surgical treatment for debilitating pain and significant loss of function, the clinician must choose between these treatment options based on clinical assessment of the patient. Furthermore, the clinician must determine the optimal values of the parameters associated with the selected treatment (e.g., method, level, and amount of correction for tibial osteotomy, and implant type, size, and positioning for joint replacement). A similar situation exists for rehabilitation and surgical treatments of neurological disorders such as stroke, Parkinson's disease, and cerebral palsy. In clinical practice, the final treatment plan is usually selected based on subjective clinical experience rather than on objective prediction of post-treatment function developed from patient data.
Personalized computational models of the neuromusculoskeletal system could facilitate objective prediction of patient-specific functional outcome for different treatment designs being considered by the clinician. Depending on the intended clinical application, a personalized neuromusculoskeletal model could account for patient-specific anatomical (e.g., skeletal structure and muscle lines of action), physiological (e.g., muscle force-generating properties), and/or neurological (e.g., constraints on achievable muscle excitation patterns) characteristics, all within the context of a multibody dynamic model. Personalized models for treatment design are motivated by the fact that for many treatments, "one size fits none." Every patient is different and possesses unique anatomical, neurological, and functional characteristics that may significantly impact optimal treatment of the patient. Personalized models provide one possible avenue for increased objectivity in treatment planning, reducing the likelihood that different clinicians will plan different treatments given the same patient data. Ideally, virtual treatments performed on a patient's personalized model would allow objective and reliable prediction of post-treatment function and thus identification of an optimal treatment plan. Such predictions would identify not only the best type of treatment (including previously unknown treatments) but also treatment parameters to which functional outcome is highly sensitive (i.e., which treatment parameter values does the clinician need to "get right"?).
This paper explores how personalized neuromusculoskeletal models could be used to improve treatment design for mobility impairments. The exploration is based on a survey of personalized modeling research being performed in Europe and thus is limited in its scope. The survey was funded by the National Science Foundation (NSF) in the United States with the goal of synthesizing research recommendations and informing research funding in the area of technology to improve mobility. In October of 2010, two teams of four panelists recruited by NSF visited a number of academic and industrial sites throughout Europe over a one week time period. Since time and financial constraints limited the number of labs that could be visited, it was not possible to gather information from all labs in Europe performing valuable work in this area. Given that the goal of the tour was to survey the state-of-the-art in Europe, we also omit discussion of valuable work being performed by labs outside of Europe. The remainder of this paper summarizes the panel's findings related to the potential clinical use and benefit of personalized neuromusculoskeletal modeling.
Clinical aspects of personalized modeling
In this section, we discuss current and future clinical uses of personalized neuromusculoskeletal models to design improved treatments for mobility impairments. To set the stage, we begin by discussing common reasons why human movement data are collected, followed by a proposal for a general process to follow when using personalized models in the treatment design process. We then discuss mobility-related clinical problems currently being addressed with personalized neuromusculoskeletal models, and we conclude this section by highlighting gaps in clinical application where personalized models could add significant value.
Clinical purpose of personalized modeling
Pre-treatment human movement (e.g., motion capture, ground reaction, muscle electromyographic, energy consumption), strength (e.g., isometric and isokinetic dynamometer), and imaging (e.g., magnetic resonance (MR), computed tomography (CT), x-ray, fluoroscopic) data provide the experimental measurements necessary to develop objective model-based predictions of post-treatment function. As described by Dr. Maria Grazia Benedetti at the Rizzoli Orthopedic Institute in Bologna, Italy, there are three primary reasons for collecting human movement data in a clinical setting:
1) it Assessment Assess after treatment how the treatment worked for an individual patient or a group of patients. An example would be using gait data to assess changes in walking speed and knee flexion angle following tendon transfer or lengthening surgery in a specific child or group of children with cerebral palsy. This use of human movement data is relatively common.
2) Identification Identify on an individual patient basis which patients should be treated (but not how they should be treated). An example would be using gait data to determine whether tendon transfer or tendon lengthening surgery should be performed for a specific child with cerebral palsy. This use of human movement data remains uncommon but is becoming more common.
3) Prediction Predict on an individual patient basis which treatment should be performed and how it should be performed. An example would be using gait data to determine whether tendon transfer or tendon lengthening surgery should be performed, which tendon to transfer or lengthen, and where to transfer it or how much to lengthen it, to improve walking ability for a specific child with cerebral palsy. This use of human movement data does not yet happen in clinical practice.
The focus of this paper is on how personalized neuromusculoskeletal models could be used for prediction rather than assessment or identification, though identification has significant clinical value as well. While prediction is the most challenging use, it is also the use with the greatest potential to improve functional outcome on an individual patient basis.
Clinical process of personalized modeling
How should personalized neuromusculoskeletal models be used to predict functional outcome for various treatment plans under consideration? Expanded from ideas presented by researchers at the Rizzoli Orthopedic Institute in Bologna, Italy, and Dr. Bart Koopman at the University of Twente in Enschede, the Netherlands, we propose a three-step process for treatment design using personalized models:
1) Model preparation steps:
indent 1 • b Identify model outputs to be used as indicators of clinical/functional outcome.
• Define model complexity required to predict these outputs with sufficient accuracy for the intended clinical application.
• Collect pre-treatment movement, strength, and imaging data (as required) to construct the personalized model and predict the outputs of interest.
2) Model construction steps:
• Calibrate model geometry and parameter values to which the outputs of interest are sensitive using pre-treatment movement, strength, and/or imaging data.
• Estimate model parameter values to which the outputs of interest are insensitive using data reported in the literature.
• Incorporate surgical or rehabilitation treatment plans under consideration into the personalized model.
3) Model utilization steps:
• Predict post-treatment patient function for each proposed treatment plan.
• Select treatment plan and associated parameter values that maximize functional outcome, possibly using numerical optimization methods.
• Validate personalized model predictions using post-treatment function measured from patients whose treatment was not planned with a model.
• Implement optimal surgical or rehabilitation treatment plan designed with the personalized model.
• Collect post-treatment movement, strength, and/or imaging data from the patient to assess clinical/functional outcome.
In this process, only the steps relevant to the mobility-related clinical application at hand need be performed. For example, clinical applications that do not require modeling of individual muscle forces may not require any imaging and strength data from the patient, and thus steps related to calibration of patient-specific muscle and bone geometry can be omitted. Model parameter values that require calibration to patient data may necessitate collection of additional experimental data solely for calibration purposes abbrgrp abbr bid B1 1.
Two critical tasks to highlight in this process are Calibrate and Validate. Unless the model is calibrated to relevant data collected from the patient prior to treatment, the model will not be sufficiently personalized to predict the patient's post-treatment function. Similarly, unless calibrated model predictions are validated using post-treatment data collected from patients whose treatments were not planned with the model, clinicians will not have confidence in the model predictions, and personalized models will never advance toward widespread clinical utility. Validation of treatment planning using personalized models will ultimately require randomized controlled trials, where outcomes are compared between patients whose treatments were planned with a personalized model and those whose treatments were not.
Clinical applications of personalized modeling
During our tour of European research labs, we sought to identify clinical applications where a personalized modeling process similar to the one outlined above was already being followed. By the end of the tour, we made three valuable observations related to clinical application of personalized neuromusculoskeletal models. First, few labs have reached the point of being able to apply this process to specific clinical problems. Second, some of the best existing clinical applications involved generic rather than personalized models. Third, most clinical applications we observed involved orthopedic surgery, with few applications involving neurorehabilitation. Below we comment further on these observations.
Three large projects funded by the European Commission (EC) are making significant strides in developing and applying personalized neuromusculoskeletal models to orthopedic clinical problems. The first is the "Osteoporotic Virtual Physiological Human" (VPHOP) project B2 2, which involves a large consortium of academic and industrial partners throughout Europe and is coordinated by Dr. Marco Viceconti at the Rizzoli Orthopedic Institute in Bologna, Italy B3 3B4 4B5 5. As stated on the project website, the goal is to "develop, validate and deploy the next generation of technology to predict the absolute risk of fracture in patients with low bone mass, thereby enabling clinicians to provide better prognoses and implement more effective treatment strategies." One of the unique emphases of the project is on multi-scale modeling, with bone being modeled simultaneously on the cell, tissue, organ, and body levels to permit clinically useful predictions of the risk of bone fracture in different patient populations.
In a related project, Viceconti's team at the Rizzoli Orthopedic Institute has developed personalized neuromusculoskeleal models of pediatric patients who received a surgical limb salvage procedure for bone cancer 4B6 6. For this clinical problem, the challenge is to determine how the patient should load the bone allograft during the rehabilitation process such that bone loads are high enough to stimulate repair but low enough to avoid fracture. Since each clinical case is unique, surgical and rehabilitation treatment design cannot be standardized. Dr. Viceconti and his research team are using gait and imaging data to create personalized neuromusculoskeletal models that estimate muscle and bone loads in the patient's femur during walking (Figure figr fid F1 1). These estimates inform the rehabilitation process and when the patient should be cleared for full functional loading with no restrictions. The primary challenges faced by this personalized modeling process are whether scaling of muscle and bone geometry from a generic model is sufficiently accurate for this pediatric application, and also whether the estimated muscle and bone loads (which currently cannot be validated experimentally) are sufficiently reliable.
fig Figure 1caption Personalized modeling workflow for massive skeletal reconstruction, as developed by the Medical Technology Laboratory at the Rizzoli Orthopedic Institute in Bologna, Italytext
Personalized modeling workflow for massive skeletal reconstruction, as developed by the Medical Technology Laboratory at the Rizzoli Orthopedic Institute in Bologna, Italy. a) CT scan of the lower limbs performed at follow-up, with motion capture markers visible as well. b) Focused view of reconstructed femur immediately after surgery. c) Patient-specific musculoskeletal model superimposed on CT images (LHPBuilder, B3C, Italy). d) One frame of dynamic walking simulation performed with the patient-specific model (OpenSim). Image courtesy of Dr. Giordano Valente, Rizzoli Orthopedic Institute, Bologna, Italy.
graphic file 1743-0003-9-18-1 hint_layout double
The second large EC-funded project is called "NMS Physiome" B7 7, which is also coordinated by researchers at the Rizzoli Orthopedic Institute. This project seeks to "promote a more organic cooperation in the development of Predictive, Personalised and Integrative musculoskeletal medicine" by integrating research efforts between the VPHOP project and the Center for Physics-based Simulation of Biological Structures (Simbios) at Stanford University in the United States. Integration is focused on neuromusculoskeletal software tools (MAF, OpenSim, and FEBio) and research community websites (BiomedTown and Simtk) developed by the two consortia. The goal of integration is to address the challenges posed by personalized neuromusculoskeletal modeling more effectively and efficiently.
The third large EC-funded project is entitled "Improving Safety and Predictability of Complex Musculo-skeletal Surgery using a Patient-Specific Navigation System" (TLEMsafe) B8 8, which involves a consortium of academic and industrial partners headed by the University of Twente in Enschede, the Netherlands. The stated goal of the project is to "create an ICT-based patient-specific surgical navigation system that helps the surgeon safely reach the optimal functional result for the patient and is a user friendly training facility for the surgeons." In this project, the researchers proposed to use personalized neuromusculoskeletal models as part of a three-step treatment design process. The first step is creation of the personalized model from the patient's movement and imaging data. The model is created within the framework of the AnyBody musculoskeletal modeling software developed by researchers at the University of Aalborg in Denmark B9 9 (Figure F2 2). The second step is for the surgeon to perform virtual surgical treatments on the personalized model and to identify the optimal surgical plan for the patient. The final step is to transfer the optimized treatment plan into a surgical navigation system to be used during actual surgery. As part of this project, Dr. Bart Koopman of the University of Twente is currently investigating the use of personalized neuromusculoskeletal models to identify optimal patient-specific tendon transfer procedures to restore hip adductor strength in patients who walk with a "drooping" swing leg hip (i.e., Trendelenburg gait due to Poliomyelitis or total hip arthroplasty).
Figure 2Examples of musculoskeletal models developed for the TLEMsafe project
Examples of musculoskeletal models developed for the TLEMsafe project. Image courtesy of Prof. Dr. Ir. Bart Koopman, University of Twente, Enschede, the Netherlands.
1743-0003-9-18-2
Other research we observed involved the use of generic rather than personalized neuromusculoskeletal models. While personalized models have the greatest potential to impact clinic practice, generic models can still provide significant clinical benefits. Two clinical applications of generic models were particularly well developed. The first was the design of a total ankle replacement by Dr. Alberto Leardini and colleagues at the Rizzoli Orthopedic Institute in Italy B10 10B11 11. The design was developed using a sagittal plane musculoskeletal model of the ankle that incorporated the ligaments and articular surfaces. The design philosophy was to maintain medial and lateral ankle ligaments in an isometric state during passive ankle motion. The team identified a novel geometric design to achieve this goal, using non-anatomically shaped tibial and talar components with a meniscal bearing interposed between them. The design has been licensed by an orthopedic implant company, and early clinical assessment is demonstrating good restoration of ankle mobility with low complication and revision rates B12 12B13 13.
The other example was evaluation of tendon transfer surgery for massive rotator cuff tears by Dr. Frans van der Helm and colleagues at Delft University of Technology in the Netherlands B14 14B15 15. The evaluation was performed using a high fidelity musculoskeletal shoulder and elbow model constructed from extensive measurements performed on a single cadaver specimen B16 16. The model accounts for more parameters (including muscle sarcomere length measured by laser diffraction) than any other upper extremity model. Simulations performed with the generic model have provided specific recommendations for which tendons to transfer, and where to transfer them, to replace the function of a torn rotator cuff without sacrificing shoulder strength for functional tasks. Similar to the pediatric oncology application, the reliability of the model's predicted clinical outcome is only as good as the reliability of the model's predicted muscle forces. Dr. van der Helm and colleagues have attempted to validate the model's prediction of shoulder muscle and contact forces using contact forces measured by an instrumented shoulder prosthesis B17 17. The authors concluded that, "Although results indicated a reasonable compatibility between model and measured data, adjustments will be necessary to individualize the generic model with the patient-specific characteristics."
The primary rehabilitation applications we observed utilized a personalized modeling method called Complementary Limb Motion Estimation (CLME) B18 18B19 19. The method uses motion measurements made on a patient's healthy leg to predict how the patient's impaired leg should move. The predictions are made in real time by exploiting the strong coupling that exists between skeletal degrees of freedom during locomotion. These couplings (or synergies) are identified in healthy subjects using statistical dimensionality reduction methods (e.g., principal component analysis) and then applied to the healthy limb of a patient to predict the desired motion of the patient's impaired limb. CLME was first proposed to generate personalized motion trajectories for the paretic limb of patients undergoing robotic gait training following stroke 19. The goal was to maintain patient stability while minimizing unwanted interaction forces between patient and robot. More recently the same method has been proposed to personalize the control of an active knee exoprosthesis to the gait patterns of patients who have undergone above-knee amputation 18. For both applications, the predicted motion of the impaired or prosthetic limb is used as a personalized reference to be tracked by the robot or prosthesis control system. While we observed other personalized rehabilitation applications, they did not have a strong neuromusculoskeletal modeling component to them.
Clinical gaps in personalized modeling
Mobility-related clinical problems are typically treated by either surgery or rehabilitation and either do or do not possess a significant neurological component. Thus, use of personalized neuromuscular models to improve treatment of mobility impairments can be grouped into four categories: 1) surgical treatment without a significant neurological component, 2) surgical treatment with a significant neurological component, 3) rehabilitation treatment without a significant neurological component, and 4) rehabilitation treatment with a significant neurological component. Of these four categories, the first is the most developed and the fourth the least developed in terms of personalized neuromusculoskeletal modeling. This situation is not surprising given that technology is a more recent addition to rehabilitation treatments (e.g., rehabilitation robotics) than to surgical treatments and that neurological factors are more difficult to model than are mechanical factors. Below we present clinical gaps in personalized modeling for each of these four categories.
Osteoarthritis is a prevalent disabling disease that is commonly treated by surgical intervention. Though it clearly possesses a neurological component B20 20, that component is secondary to mechanical factors as far as surgical treatment design is concerned. Use of personalized neuromusculoskeletal models has been proposed by European labs for pre-operative planning of high tibial osteomities and total joint replacements B21 21B22 22. While joint replacement surgery is generally reliable, individual cases can pose special challenges, especially those involving revision surgery. In contrast, high tibial osteotomy (HTO) is a challenging surgical procedure with highly variable outcomes but also high potential benefits, making it an excellent target for personalized models. Use of personalized models to design customized gait modifications following HTO surgery could also be valuable for avoiding the loss of boney correction that often occurs over time. Anterior cruciate ligament replacement to avoid knee osteoarthritis is a related surgical application where personalized models could be of value B23 23.
In contrast, cerebral palsy and Charcot-Marie-Tooth disease are neurological disorders that are commonly treated by surgery, since no treatment exists for the underlying neurological problem. Though Charcot-Marie-Tooth disease is not well known, it is the most commonly inherited neurological disorder and limits mobility in approximately 1 in 2,500 individuals B24 24. Surgical treatments for both disorders typically involve muscle lengthening, tendon transfer, and/or osteotomy to improve joint range of motion, foot-ground contact pattern, gait speed, and gait symmetry. For both disorders, patients have varied and unique clinical presentations, making stereotypical treatment planning ineffective. For this reason, personalized neuromusculoskeletal models, especially those that are able to model the neurological limitations (e.g., muscle spasticity) of the patient, could play a valuable role in predicting the outcome of complex multi-level surgeries that are performed on these patients B25 25B26 26B27 27.
Stroke, spinal cord injury, and traumatic brain injury significantly affect mobility, possess a major neurological component, and are often treated by rehabilitation methods. Personalized neuromusculoskeletal models have yet to be applied to traditional or robot-assisted rehabilitation treatments for these disorders. Given this large gap, even personalized models that omit neural control models have the potential to make a significant clinical impact. For example, a number of clinical and research labs in Europe are utilizing robot-assisted therapy for neurorehabilitation B28 28B29 29. Many of these labs are using the Lokomat gait trainer (Hocoma AG, Volketswil, Switzerland), whose programmed walking pattern is that of one of the designers. Personalized musculoskeletal models that can predict patient-specific improvements in gait pattern could be used to customize robot-prescribed gait motions. For individuals who have had a stroke, similar models could be used to predict a patient-specific sequence of gradual gait alterations leading to normal function, with the model indentifying where to focus rehabilitation efforts to maximize functional outcome.
Personalized musculoskeletal models that include personalized neural control models would be even more beneficial for improving rehabilitation of neurological disorders. Models that account for patient-specific neural control limitations and neuroplasticity could be used to identify the maximum expected improvement and how best to get there. As suggested by Dr. Herman van der Kooij of the University of Twente in the Netherlands, such models could be useful for predicting how people interact with and adapt to their environments, which could improve the effectiveness of robotic therapy systems. Furthermore, such models could be valuable for the design of neuroprostheses that use functional electrical stimulation to restore lost function B30 30B31 31. For example, if a personalized neuromusculoskeletal model could predict a minimum set of muscles to stimulate, and how and when to stimulate them, to restore a normal gait pattern, then the personalized prescription could be investigated in a clinical environment. As stated by one of the researchers on our panel, "There is a need for. . improved models of human motor recovery to provide a more rational framework for designing robotic therapy control strategies." B32 32.
One of the primary reasons for these clinical gaps is lack of effective collaboration between clinical researchers and personalized modeling researchers. An excellent counterexample is the Rizzoli Orthopedic Institute in Italy, where clinicians and engineers share the same office space and interact during clinical decision making. These interactions create an atmosphere where clinicians routinely enter into the technical world and engineers routinely enter into the clinical world. Such an environment of shared intellectual investment in solving clinical problems is critical if personalized neuromusculoskeletal modeling is to make a broad impact in the clinic.
Technical aspects of personalized modeling
Significant research efforts are currently underway in labs throughout Europe to develop personalized neuromusculoskeletal modeling tools and methods. Recalling the section above on the Clinical Process of Personalized Modeling, the primary challenges faced by these efforts are the Calibrate step within Model construction and the Validate step within Model utilization. In this section, we discuss current technical capabilities of personalized modeling related to model calibration and validation, followed by a discussion of technical gaps that need to be filled if personalized neuromusculoskeletal models are to become clinically useful.
Technical capabilities of personalized modeling
Despite significant computational advances over the past ten years, model personalization remains a major challenge, as does the ability to use a personalized model to predict the outcome of a clinical intervention. Personalized neuromusculoskeletal models can be applied to mobility-related clinical problems only to the extent to which key model features can be calibrated to data collected from a patient. Thus, the ability to calibrate models to patient data is a prerequisite to clinical use of personalized models, with the proposed clinical application determining the extent of model personalization required.
Since most neuromusculoskeletal models are generic, being constructed from detailed anatomic measurements performed on cadaver specimens 16B33 33, a model personalization (or calibration) process is needed. Expanding on information provided by Dr. Bart Koopman of the University of Twente in the Netherlands, we propose four model calibration steps that should be performed in whole or in part to transform a generic model into a personalized model:
1) Geometric calibration Use of imaging data (e.g., MR, CT, x-ray) to calibrate bone geometry, muscle lines of action, and muscle moment arms in a musculoskeletal model.
2) Kinematic calibration Use of motion data (e.g., marker-based, inertial sensors, fluoroscopy) to calibrate constraint-based joint positions and orientations in the body segments of a skeletal model.
3) Kinetic calibration Use of load data (e.g., ground reaction force and moment, foot contact pressure, dynamometer) to calibrate body segment mass and inertia, foot stiffness, muscle strength, and other muscle-tendon properties in a neuromusculoskeletal model.
4) Neurologic calibration Use of motion, load, and muscle activity data (i.e., muscle EMG) to calibrate feedforward, intrinsic feedback, reflexive feedback, and/or synergy properties of the neural control system in a neuromusculoskeletal model.
The challenge is how to construct a personalized model that is consistent with all available data from these different modalities B34 34.
Current methods for geometric calibration involve uniform scaling, non-uniform scaling, deformation, or direct creation of bone models and muscle lines of action from patient MR or CT data. Uniform scaling based on external measurements is inaccurate when calculating muscle moment arms, muscle-tendon lengths, muscle forces, and joint contact forces B35 35, especially when scaling a generic model of an adult to a pediatric patient B36 36. Non-uniform scaling is only slightly better at producing accurate muscle moment arms and muscle-tendon lengths 27. Creation of patient-specific geometry directly from the patient's imaging data remains the gold standard 26, but the process is highly time consuming and somewhat subjective, depending on the imaging modality and the anatomic structures being modeled (e.g., bone edges are often poorly defined in MR data).
Kinematic calibration involves determining fixed joint positions and orientations in the body segments of a skeletal model with a pre-defined kinematic structure. The calibration process is usually performed using surface marker data, with joint angles in the model being calculated as a byproduct. European labs have used optimization methods 1 and extended Kalman filter methods B37 37B38 38B39 39 to perform kinematic calibration. Filter-based methods have the advantage of being computationally faster and less complex than most optimization methods, but they require a greater amount of algorithm tuning to achieve satisfactory performance. Despite these advances, neither approach has yet to be generalized and incorporated into commercial-grade musculoskeletal modeling software such as the AnyBody program.
Kinetic calibration typically involves calibration of segment mass properties or muscle force-generating properties. Though segment mass properties can be calibrated to force plate and motion data B40 40, they are often taken from regression equations developed from measurements performed on cadavers B41 41. Similarly, though muscle model parameter values (e.g., muscle strength) can be calibrated to isometric and/or isokinetic dynamometer data B42 42, this calibration step is usually omitted due to the extra effort it requires. A recent European study indicated that subject-specific muscle-tendon parameter values calibrated to dynamometer data are appropriate for use in musculoskeletal models used to analyze gait B43 43. Since many movement impairments involve undesirable foot-ground contact patterns, kinetic calibration of patient-specific foot-ground contact models will be essential in the future for predicting changes in gait function due to various proposed treatments, yet kinetic calibration methods for such models do not yet exist.
A promising development to address geometric, kinematic, and kinetic calibration simultaneously is research being performed by Dr. Wafi Skalli at Arts et Métiers ParisTech in Paris, France using the EOS bi-plane x-ray system (EOS Imaging SA, Paris, France). With the EOS system, a subject stands upright while a low-dose x-ray system scans the entire body from head to toe collecting one continuous distortion-free image in each of two orthogonal planes. The resulting bi-plane images are then processed and morphed using template anatomy for personalization and visualization. Geometric calibration of bone and muscle geometry via deformation of template bone and muscle models can be performed rapidly and accurately relative to geometry constructed directly from CT data B44 44B45 45B46 46B47 47. Kinematic calibration could theoretically be aided by performing scans of the relevant portion of the body in two or more poses (e.g., the lower extremities in different squatting positions) B48 48, especially if surface markers to be used in additional movement experiments are also visible. Kinetic calibration of segment mass properties and muscle strength parameters (based on muscle cross sectional areas) can also be performed from the images B49 49B50 50. Thus, with improvements in automation and refinement of existing algorithms, the EOS system has the potential to improve musculoskeletal model personalization significantly.
The remaining area, neurologic calibration, has seen the least progress due to the significant challenges involved in understanding how the human neural control system functions. This calibration step can be pursued using a range of approaches, from a physiological approach that seeks to model the detailed anatomy and physiology involved in neural control, to an emergent approach that seeks to model the neural control computations implemented by the anatomy but without modeling anatomic detail. An example of a physiological approach is detailed modeling of feedforward, intrinsic (i.e., muscle) feedback, and reflexive (i.e., visual, proprioceptive, and vesitibular) feedback mechanisms utilized by the neural control system B51 51B52 52B53 53B54 54. To date, such high fidelity neural control models have been applied to postural control rather than movement tasks, and methods for personalizing the parameter values in these models are not yet well developed. At the other extreme, an example of an emergent approach is muscle synergy analysis, where EMG signals from a large number of muscles (e.g., 16) are decomposed into a smaller number of basis activation signals (e.g., 5) for all muscles plus a unique set of weights (often termed "modules") for each muscle that scale the activation signals B55 55B56 56. Synergy analysis is used for dimensionality reduction (e.g., 5 basis signals are used to reconstruct 16 EMG signals) and can identify neural control limitations in patients following stroke B57 57. Incorporation of these limitations into personalized neuromusculoskeletal models could facilitate prediction of best possible functional outcome. Between these two extremes is a physiological approach that has successfully explained motor learning using simplified feedforward and feedback models B58 58. The approach uses a v-shaped learning function to model the change in muscle feedforward commands generated in response to kinematic errors experienced during the previous movement trial. Computer simulation of a sequence of arm movement trials performed in different force fields revealed that the method can successfully reproduce experimentally observed trial-to-trial changes in muscle activations (to control force) and co-contraction (to control impedence).
Validation of clinical predictions is the other major challenge faced by personalized models. This challenge can be surmounted when clinical outcome variables are external quantities that can be easily measured (e.g., gait speed, gait symmetry). Frequently, however, established clinical databases use coarse ordinal scales to rank movement ability, and mapping these scores to neuromechanical models is difficult. In addition, significant challenges remain when the outcome variables are either internal to the body (e.g., muscle forces, joint contact forces, bone strains) or dependent on quantities that are internal to the body. Since such quantities cannot be measured directly by non-invasive means, alternate methods are needed for personalized model validation. For example, predicted muscle forces have been evaluated indirectly using in vivo joint contact force measurements 17 or novel measurements (e.g., near infrared spectroscopy) that are likely to be highly correlated with in vivo muscle force B59 59.
Technical gaps in personalized modeling
As suggested by this review of current technical capabilities in Europe, at least four critical technical gaps currently exist that limit the potential clinical applicability of personalized neuromusculoskeletal models:
1) How can we make the personalized model calibration and prediction process fast and easy?
Though several excellent musculoskeletal modeling programs exist, none of them contain functionality that automates the model calibration process and simplifies the model prediction process. Personalized model calibration and prediction currently require significant expertise and programming ability possessed by only a small number of researchers in a limited number of research labs. Making these capabilities available to the larger neuromusculoskeletal modeling community via fast, automated algorithms will be essential for the growth of personalized modeling efforts. Ultimately, personalized modeling will be adopted for routine clinical use only when it is extremely easy to use.
2) How can we calibrate "unobservable" parameters to which model predictions are sensitive?
For some clinical problems, personalized model predictions of functional outcome will be sensitive to model parameter values that cannot be calibrated to available data. The first step in addressing this problem is identifying when it occurs, which requires performing sensitivity analyses that in some cases will be limited by existing computational capabilities. The next step is development of new experimental methods or hardware that provide sufficiently rich information to calibrate the parameter values needed to develop the predictions.
3) How can we create personalized neural control models?
Few neuromusculoskeletal models published to date account for any level of personalized neural control modeling. Such modeling would ideally account for limitations in a patient's neural control capabilities as well as the extent of possible plasticity. Emergent approaches for modeling neural control could be incorporated into personalized musculoskeletal models as a starting point, while physiological models could be refined to the point where essential model parameters become well defined and methods for calibrating them are developed. The ability to incorporate complex personalized neural control models into personalized musculoskeletal models would greatly expand model applicability to clinical situations, especially those involving neurorehabilitation.
4) How can we validate model-based predictions, especially for internal quantities such as muscle, joint, and bone loads?
Validation of internal quantities that influence treatment design remains a major challenge. While researchers continue to refine optimization and EMG-driven methods for predicting muscle forces and related joint and bone loads, the ability to validate these predictions has lagged behind. Direct measurement of internal quantities under special conditions (e.g., instrumented implants), and the opportunity to test model-based predictions against these internal measurements, provides a valuable avenue for model validation efforts B60 60. Identification of novel approaches that utilize only existing data collection capabilities, as well as development of new experimental techniques, will be essential if clinicians are to gain confidence in treatment plans designed with personalized neuromusculoskeletal models.
Conclusions
Neuromusculoskeletal modeling has yet to make a significant difference in routine clinical practice. For this situation to change, the key gaps identified above need to be addressed by modeling researchers in close collaboration with clinical investigators. While the biggest clinical gap for personalized neuromusculoskeletal modeling is in neurorehabilitation, the gap for other mobility-related clinical problems is almost as large. The biggest technical gap is in personalized neural control and recovery models, though issues like automation of the model personalization process and development of personalized foot-ground contact models are critical as well for advancement. For clinical problems that involve highly unique patient characteristics, stereotypical treatment design is likely to yield variable functional outcomes. These types of clinical problems are where personalized neuromusculoskeletal models have the greatest potential to create a positive paradigm shift in the treatment design process.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
BF organized and drafted the manuscript based on information gathered during the European tour. ML provided clinical perspective on information gathered. DR organized the tour and the selection of European sites. All authors made significant contributions to gathering, critically evaluating, organizing, and revising the content of the manuscript. All authors read and approved the final manuscript.
bm
ack
Acknowledgements
The authors wish to thank Dr. Ted Conway of the National Science Foundation for funding this review of European research and World Technology Evaluation Center for managing the logistics of the study tour.
refgrp A computationally efficient optimisation-based method for parameter identification of kinematically determinate and over-determinate biomechanical systemsAndersenMSDamsgaardMMacWilliamsBRasmussenJComput Methods Biomech Biomed Eng201013171lpage 18310.1080/10255840903067080Osteoporotic Virtual Physiological Human Projecthttp://www.vphop.eu/The virtual physiological human: computer simulation for integrative biomedicine IVicecontiMKohlPPhil Trans A Math Phys Eng Sci20103682591259410.1098/rsta.2010.0096Multiscale modelling of the skeleton for the prediction of the risk of fractureVicecontiMTaddeiFVan Sint JanSLeardiniACristofoliniLSteaSBaruffaldiFBaleaniMClin Biomech20082384585210.1016/j.clinbiomech.2008.01.009The Virtual Physiological Human a European initiative for in silico human modellingVicecontiMClapworthyGVan Sint JanSJ Physiol Sci20085844144610.2170/physiolsci.RP009908link fulltext 18928640Femoral loads during gait in a patient with massive skeletal reconstructionTaddeiFMartelliSValenteGLeardiniABenedettiMGManfriniMVicecontiMClin Biomech2011(conditionally accepted)NMS Physiome Projecthttp://www.nmsphysiome.eu/Improving Safety and Predictability of Complex Musculo-skeletal Surgery using a Patient-Specific Navigation System (TLEMsafe Project)http://www.tlemsafe.eu/Analysis of musculoskeletal systems in the AnyBody modeling systemDamsgaardMRasmussenJChristensenSTSurmaEde ZeeJSimul Model Prac Theory2006141100111110.1016/j.simpat.2006.09.001Computer-assisted design of the sagittal shapes of a ligament-compatible total ankle replacementLeardiniACataniFGianniniSO'ConnorJJMed Biol Eng Comput20013916817510.1007/BF0234479911361242Mobility of the human ankle and the design of total ankle replacementLeardiniAO'ConnorJJCataniFGianniniSClin Orthop Relat Res200442439615241142Gait analysis in patients operated with a novel total ankle prosthesisIngrossoSBenedettiMGLeardiniACasanelliSSforzaTGianniniSGait Posture20093013213710.1016/j.gaitpost.2009.03.01219477648Total ankle replacement compatible with ligament function produces mobility, good clinical scores, and low complication rates: an early clinical assessmentGianniniSRomagnoliMO'ConnorJJMalerbaFLeardiniAClin Orthop Relat Res20104682746275310.1007/s11999-010-1432-3pmcid 304963120559763Biomechanical analysis of tendon transfers for massive rotator cuff tearsMagermansDJChadwickEKVeegerHEvan der HelmFCRozingPMClin Biomech20041935035710.1016/j.clinbiomech.2003.11.013Effectiveness of tendon transfers for massive rotator cuff tears: a simulation studyMagermansDJChadwickEKVeegerHERozingPMvan der HelmFCClin Biomech20041911612210.1016/j.clinbiomech.2003.09.008Measuring muscle and joint geometry parameters of a shoulder for modeling purposesKlein BretelerMDSpoorCWVan der HelmFCJ Biomech1999321191119710.1016/S0021-9290(99)00122-010541069Validation of the Delft Shoulder and Elbow Model using in-vivo glenohumeral joint contact forcesNikooyanAAVeegerHEWesterhoffPGraichenFBergmannGvan der HelmFCJ Biomech2010433007301410.1016/j.jbiomech.2010.06.01520655049Complementary limb motion estimation for the control of active knee prosthesesValleryHBurgkartRHartmannCMitternachtJRienerRBussMBiomed Tech (Berl)2011564551Reference trajectory generation for rehabilitation robots: complementary limb motion estimationValleryHvan AsseldonkEHBussMvan der KooijHIEEE Trans Neural Syst Rehabil Eng200917233019211320A neuromuscular mechanism of posttraumatic osteoarthritis associated with ACL injuryPalmieri-SmithRMThomasACExerc Sport Sci Rev20093714715310.1097/JES.0b013e3181aa666919550206Musculoskeletal biomechanics of the knee joint. Principles of preoperative planning for osteotomy and joint replacementHellerMOMatziolisGKonigCTaylorWRHinterwimmerSGraichenHHegeHCBergmannGPerkaCDudaGNOrthopade20073662863410.1007/s00132-007-1115-217605127Musculoskeletal load analysis. A biomechanical explanation for clinical results--and more?HellerMOSchroderJHMatziolisGSharenkovATaylorWRPerkaCDudaGNOrthopade20073618819410.1007/s00132-007-1054-y17333071Elastic properties of an intact and ACL-ruptured knee joint: measurement, mathematical modelling, and haptic renderingFreyMRienerRMichasCRegenfelderFBurgkartRJ Biomech2006391371138210.1016/j.jbiomech.2005.04.02116039659Charcot-Marie-Tooth diseaseReillyMMMurphySMLauraMJ Peripher Nerv Syst20111611421696489Validation of hamstrings musculoskeletal modeling by calculating peak hamstrings length at different hip anglesvan der KrogtMMDoorenboschCAHarlaarJJ Biomech2008411022102810.1016/j.jbiomech.2007.12.01018222456Level of subject-specific detail in musculoskeletal models affects hip moment arm length calculation during gait in pediatric subjects with increased femoral anteversionScheysLDesloovereKSuetensPJonkersIJ Biomech2011441346135310.1016/j.jbiomech.2011.01.00121295307Calculated moment-arm and muscle-tendon lengths during gait differ substantially using MR based versus rescaled generic lower-limb musculoskeletal modelsScheysLSpaepenASuetensPJonkersIGait Posture20082864064810.1016/j.gaitpost.2008.04.01018534855Design and evaluation of the LOPES exoskeleton robot for interactive gait rehabilitationVenemanJFKruidhofRHekmanEEEkkelenkampRVan AsseldonkEHvan der KooijHIEEE Trans Neural Syst Rehabil Eng20071537938617894270Patient-cooperative control increases active participation of individuals with SCI during robot-aided gait trainingDuschau-WickeACaprezARienerRJ Neuroeng Rehabil201074310.1186/1743-0003-7-43294970720828422Model-based development of neuroprosthesis for paraplegic patientsRienerRPhilos Trans R Soc Lond B Biol Sci199935487789410.1098/rstb.1999.0440169258710382222Model-based development of neuroprostheses for restoring proximal arm functionKirschRFAcostaAMvan der HelmFCRotteveelRJCashLAJ Rehabil Res Dev20013861962611767969Review of control strategies for robotic movement training after neurologic injuryMarchal-CrespoLReinkensmeyerDJJ Neuroeng Rehabil200962010.1186/1743-0003-6-20271033319531254Morphological muscle and joint parameters for musculoskeletal modelling of the lower extremityKlein HorsmanMDKoopmanHFvan der HelmFCProseLPVeegerHEClin Biomech20072223924710.1016/j.clinbiomech.2006.10.003A new software tool for 3D motion analyses of the musculo-skeletal systemLeardiniABelvedereCAstolfiLFantozziSVicecontiMTaddeiFEnsiniABenedettiMGCataniFClin Biomech20062187087910.1016/j.clinbiomech.2006.03.007Subject-specific hip geometry and hip joint centre location affects calculated contact forces at the hip during gaitLenaertsGBartelsWGelaudeFMulierMSpaepenAVan der PerreGJonkersIJ Biomech2009421246125110.1016/j.jbiomech.2009.03.03719464012Personalized MR-based musculoskeletal models compared to rescaled generic models in the presence of increased femoral anteversion: effect on hip moment arm lengthsScheysLVan CampenhoutASpaepenASuetensPJonkersIGait Posture20082835836510.1016/j.gaitpost.2008.05.00218571416Kinematical models to reduce the effect of skin artifacts on marker-based human motion estimationCerveriPPedottiAFerrignoGJ Biomech2005382228223610.1016/j.jbiomech.2004.09.03216154410Kalman smoothing improves the estimation of joint kinematics and kinetics in marker-based human gait analysisDe GrooteFDe LaetTJonkersIDe SchutterJJ Biomech2008413390339810.1016/j.jbiomech.2008.09.03519026414Tracking the motion of hidden segments using kinematic constraints and Kalman filteringHalvorsenKJohnstonCBackWStokesVLanshammarHJ Biomech Eng200813001101210.1115/1.283803518298188Influence of body segment parameters and modeling assumptions on the estimate of center of mass trajectoryLenziDCappelloAChiariLJ Biomech2003361335134110.1016/S0021-9290(03)00151-912893042Influence of body segments' parameters estimation models on inverse dynamics solutions during gaitRaoGAmarantiniDBertonEFavierDJ Biomech2006391531153610.1016/j.jbiomech.2005.04.01415970198Calibration of EMG to force for knee muscles is applicable with submaximal voluntary contractionsDoorenboschCAJoostenAHarlaarJJ Electromyogr Kinesiol20051542943510.1016/j.jelekin.2004.11.00415811613Sensitivity of dynamic simulations of gait and dynamometer experiments to hill muscle model parameters of knee flexors and extensorsDe GrooteFVan CampenAJonkersIDe SchutterJJ Biomech2010431876188310.1016/j.jbiomech.2010.03.02220392450Volumic patient-specific reconstruction of muscular system based on a reduced dataset of medical imagesJolivetEDaguetEPomeroVBonneauDLaredoJDSkalliWComput Methods Biomech Biomed Eng20081128129010.1080/102558408019594793D reconstruction of the spine from biplanar X-rays using parametric models based on transversal and longitudinal inferencesHumbertLDe GuiseJAAubertBGodboutBSkalliWMed Eng Phys20093168168710.1016/j.medengphy.2009.01.003192307433D reconstruction of the pelvis from bi-planar radiographyMittonDDeschenesSLaporteSGodboutBBertrandSde GuiseJASkalliWComput Methods Biomech Biomed Eng200691510.1080/10255840500521786Fast 3D reconstruction of the lower limb using a parametric model and statistical inferences and clinical measurements calculation from biplanar X-raysChaibiYCressonTAubertBHausselleJNeyretPHaugerOde GuiseJASkalliWComput Methods Biomech Biomed Eng2011inpress Tibio-femoral joint constraints for bone pose estimation during movement using multi-body optimizationBergaminiEPilletHHausselleJThoreuxPGuerardSCamomillaVCappozzoASkalliWGait Posture20113370671110.1016/j.gaitpost.2011.03.00621458992Personalized body segment parameters from biplanar low-dose radiographyDumasRAissaouiRMittonDSkalliWde GuiseJAIEEE Trans Biomed Eng2005521756176310.1109/TBME.2005.85571116235661Subject-specific body segment parameters' estimation using biplanar X-rays: a feasibility studySandozBLaporteSSkalliWMittonDComput Methods Biomech Biomed Eng20101364965410.1080/10255841003717608Non-linear stimulus-response behavior of the human stance control system is predicted by optimization of a system with sensory and motor noisevan der KooijHPeterkaRJJ Comput Neurosci2011An adaptive model of sensory integration in a dynamic environment applied to human stance controlvan der KooijHJacobsRKoopmanBvan der HelmFBiol Cybern20018410311510.1007/s00422000019611205347Quantifying proprioceptive reflexes during position control of the human armSchoutenACde VlugtEvan HiltenJJvan der HelmFCIEEE Trans Biomed Eng20085531132118232375Analysis of reflex modulation with a biologically realistic neural networkStienenAHSchoutenACSchuurmansJvan der HelmFCJ Comput Neurosci20072333334810.1007/s10827-007-0037-7279962417503169Five basic muscle activation patterns account for muscle activity during human locomotionIvanenkoYPPoppeleRELacquanitiFJ Physiol2004556267282166489714724214Motor control programs and walkingIvanenkoYPPoppeleRELacquanitiFNeuroscientist20061233934810.1177/107385840628798716840710Impulses of activation but not motor modules are preserved in the locomotion of subacute stroke patientsGizziLNielsenJFFeliciFIvanenkoYPFarinaDJ Neurophysiol2011CNS learns stable, accurate, and efficient movements using a simple algorithmFranklinDWBurdetETeeKPOsuRChewCMMilnerTEKawatoMJ Neurosci200828111651117310.1523/JNEUROSCI.3099-08.200818971459The relationship between two different mechanical cost functions and muscle oxygen consumptionPraagmanMChadwickEKvan der HelmFCVeegerHEJ Biomech20063975876510.1016/j.jbiomech.2004.11.03416439246Grand Challenge Competition to Predict In Vivo Knee LoadsFreglyBJBesierTFLloydDGDelpSLBanksSAPandyMGD'LimaDDJ Orthop Res20123050351310.1002/jor.2202322161745