Walking-adaptability assessments with the Interactive Walkway: Between-systems agreement and sensitivity to task and subject variations

Walking-adaptability assessments with the Interactive Walkway:

Between-systems agreement and sensitivity to task and subject variations

DaphneJ.Geerse^a,b,*,Bert H.Coolen^a,MelvynRoerdink^a

aDepartmentofHumanMovementSciences,FacultyofBehaviouralandMovementSciences,VrijeUniversiteitAmsterdam,AmsterdamMovementSciences, TheNetherlands

bDepartmentofNeurology,LeidenUniversityMedicalCenter,Leiden,TheNetherlands

1. Introduction

Animportantaspectofwalkingisone’sabilitytoadaptwalking to environmental circumstances [1–3]. Walking adaptability includestheabilityto avoid obstacles,make suddenstops and starts and accurately place the feet to environmental context [1].Mostwalking-relatedfallsresultfrominadequateinteractions withenvironmentalcontext,leadingtobalancelossduetoatrip,

slipormisplacedstep[4–6].Walkingadaptabilitythusseemstobe animportantdeterminantoffallrisk,yetacomprehensivewell- testedobjectiveassessmentofwalkingadaptabilityislacking[1].

WetrytofillthislacunawiththeInteractiveWalkway(IWW),a 10-mwalkwayaugmentedwithprojectedgait-dependentvisual context,suchasobstaclessuddenlyappearingatthepositionone would step next, demanding a step adjustment under time pressure.ThebasisoftheIWWisanintegratedmulti-Kinectv2 set-up for markerless registration of 3D full-body kinematics duringwalking[7],whichwasrecentlyvalidatedovertheentire 10-m walkway against a gold standard in 3D measurement accuracyforbothkinematicsandderivedgaitparameters[7,8].We havenowequippedthisset-upwithaprojectortoaugmentthe entirewalkway withvisualcontext,suchas obstacles,sudden- ARTICLE INFO

Articlehistory:

Received1September2016

Receivedinrevisedform15January2017 Accepted21February2017

Keywords:

Kinectv2

Walkingadaptability Assessment

Between-systemsagreement Sensitivitytotaskvariation Sensitivitytosubjectvariation

ABSTRACT

Theabilitytoadaptwalkingtoenvironmentalcircumstancesisanimportantaspectofwalking,yet difficulttoassess.TheInteractiveWalkwaywasdevelopedtoassesswalkingadaptabilitybyaugmenting amulti-Kinect-v210-mwalkwaywithgait-dependentvisualcontext(steppingtargets,obstacles)using real-time processed markerless full-body kinematics. In this study we determined Interactive Walkway’susabilityforwalking-adaptabilityassessmentsin termsofbetween-systemsagreement andsensitivity to task andsubject variations.Undervarying taskconstraints, 21healthysubjects performed obstacle-avoidance, sudden-stops-and-starts and goal-directed-stepping tasks. Various continuouswalking-adaptabilityoutcomemeasureswereconcurrentlydeterminedwiththeInteractive Walkwayandagold-standardmotion-registrationsystem:availableresponsetime,obstacle-avoidance and sudden-stop margins, steplength, stepping accuracy and walking speed.The sameholds for dichotomousclassificationsofsuccessandfailureforobstacle-avoidanceandsudden-stopstasksand performedshort-strideversuslong-strideobstacle-avoidancestrategies.Continuouswalking-adapt- ability outcome measures generally agreed well between systems (high intraclass correlation coefficientsforabsoluteagreement, low biasesandnarrowlimits ofagreement) and werehighly sensitivetotaskandsubjectvariations.Successandfailureratingsvariedwithavailableresponsetimes andobstacle typesand agreedbetweensystemsfor85–96%ofthetrialswhileobstacle-avoidance strategieswerealwaysclassifiedcorrectly.WeconcludethatInteractiveWalkwaywalking-adaptability outcomemeasuresarereliableandsensitivetotaskandsubjectvariations,eveninhigh-functioning subjects. We therefore deem Interactive Walkway walking-adaptability assessments usable for obtaininganobjectiveandmore task-specificexaminationof one’sabilityto walk, whichmaybe feasibleforbothhigh-functioningandfragilepopulationssincewalkingadaptabilitycanbeassessedat variouslevelsofdifficulty.

ß2017ElsevierB.V.Allrightsreserved.

* Correspondingauthorat:VrijeUniversiteitAmsterdam,DepartmentofHuman Movement Sciences, Van der Boechorststraat 7, 1081 BT Amsterdam, The Netherlands.

E-mailaddress:d.j.geerse@vu.nl(D.J.Geerse).

ContentslistsavailableatScienceDirect

Gait & Posture

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

http://dx.doi.org/10.1016/j.gaitpost.2017.02.021 0966-6362/ß2017ElsevierB.V.Allrightsreserved.

stop-and-start cues and stepping targets, based on real-time processedintegratedKinectdata.Theso-elicitedgait-environment interactions potentially allow for assessing various walking- adaptabilityaspects(e.g.,theabilitytoavoidobstacles,suddenly stop or start, performaccurate goal-directed steps) as well as subject-specific variations and adaptations affecting walking- adaptabilityperformance(e.g.,adoptingaslower walkingspeed toenhancegoal-directedsteppingaccuracy).

Theobjectiveofthisstudyistodeterminetheusabilityofthe IWWforwalking-adaptabilityassessmentsinagroupofhealthy adultsintermsofbetween-systemsagreementandsensitivityto task and subject variations. Walking-adaptability tasks and associatedoutcomemeasuresareselectedfortheirprovenability to distinguish between persons who vary in adaptive-walking limitations[2,3,9–12].Todeterminethe between-systemsagree- ment, IWW-based walking-adaptability outcome measures are comparedtothoseconcurrentlyderivedwithagoldstandard.The sensitivity to task variation is assessed by comparing walking- adaptability performance as a function of context variations, includingdifferentobstaclesizesandsequencesofsteppingtargets.

Sensitivitytosubject variationisexplored byquantifyingspeed- performancetrade-offsbetweenself-selected walkingspeedand adaptivesteppingperformance(successrates,safetymargins).We expect that walking-adaptability outcomes agree well between systemsandaresensitivetotaskandsubjectvariations.

2. Methods 2.1. Subjects

Aheterogeneousgroupof21healthysubjects(mean[range]:

age30[19–63]years,height176[158–190]cm,weight70[53–

83]kg, 11 males) withoutseverevisualdeficitsor anymedical conditionthatwouldaffectwalkingparticipated.Thelocalethics committeeapprovedthestudy.Allsubjectsgavewritteninformed consentpriortoparticipation.

2.2. Experimentalset-upandprocedure

Full-bodykinematicsforwalkingovertheentire10-mwalkway wasobtainedwiththeIWWusingfourspatiallyandtemporally integratedKinect v2sensors (Fig.1A)andtheOptotrak system (Northern Digital Inc., Waterloo, Canada) for 19 matched body points as in [7; see also Supplementary material]. IWW and Optotrak data were sampled at 30Hz (using custom-written softwareutilizingtheKinect-for-WindowsSoftwareDevelopment Kit[SDK2.0]) and60Hz(usingFirst Principlesdataacquisition software),respectively.TheIWWwasequippedwithaprojector (Vivitek D7180HD, ultra-short-throw Full HD projector) to augmenttheentire10-mwalkwaywithvisualcontextforthree sortsof walking-adaptability tasks:obstacle avoidance, sudden stops-and-startsandgoal-directedstepping(Fig.1).

Theobstacle-avoidancetaskconsistedof25trialswithoneor two obstacles (a projected red rectangle) per trial. In total, 40 obstacles were presented, including 20 gait-dependent obstacles(obstacleatpredictedfoot-placementpositionappearing two stepsahead; Fig.1B) and 20 position-dependent obstacles (obstacleatanunpredictablepredefinedpositionappearingwhen a subject’sankle waswithin1.5mfromthatobstacle; Fig.1C).

Gait-dependent obstacles were 0.5 (width of the walkway) by 0.3m.Position-dependentobstacleswerelarger(0.5m0.5m)to increase the need for making step adjustments. Subjects were instructedtoavoidsuddenlyappearingobstacleswhilewalkingat self-selectedcomfortablespeeds.

Thesudden-stops-and-startstask(Fig.1D)consistedof25trials withintotal40cues(i.e.,oneortwosudden-stop-and-startcues

pertrial)toassessone’sabilitytosuddenlystopandstartwalking.

Thecuewasabigbluerectanglewithawidthof0.5mthatfilled thewalkwayfromanunpredictablepredefinedpositiontillitsend andappearedassoonasasubject’sanklewaswithin1mfromthis position, triggeringthesubjecttostopwalking.Aftera random periodbetween5and10s,therectangledisappeared,triggering thesubjecttostartwalkingagain.Subjectswereinstructedtowalk atself-selectedcomfortablespeedsandtostopbehindthecueand tostartwalkingassoonasthecuedisappeared.

The goal-directed-stepping task consistedof symmetric-step- ping-stones(SSS;Fig.1E),asymmetric-stepping-stones(ASS;Fig.1F) and variable-stepping-stones (VSS;Fig. 1G) conditions. Subjects were instructedtostepasaccuratelyaspossibleontothewhiteshoe-size- matched steppingtargets ata self-selectedcomfortable walking speed.ForSSS,sevendifferentimposedstep-lengthtrialsranging from30to90cminstepsof10cmwereperformed,allwiththree repetitions, yielding a total of 21 trials. For ASS, stride length remained 90cmwhileleft(L)andright(R)imposedsteplengthswerevaried inseparatetrialsfrom15to75cminstepsof15cmyieldingfive differentimposedsteppingasymmetries(L/R:15/75,30/60,45/45, 60/30,75/15),allwiththreerepetitions,yielding15trials.ForVSS, imposedsteplengthsvariedwithineachtrialonastep-to-stepbasis randomly between30 and 90cm. Ten differentVSS trials were performed,consistingof21steppingstoneseach.

The walking-adaptability tasks were block-randomized and precededbyafamiliarizationtrial.Fourankle-to-shoecalibration trials,inwhichthesubjectwasstandingintwoshoe-size-matched

Fig.1.Theset-upoftheInteractiveWalkwaywithvisualcontextprojectedonthe walkway(A).ThefourKinectv2sensorswerepositionedontripodsataheightof 0.75malongsideawalkwayof10by0.5m.Thesensorswereplacedfrontoparallel (i.e.,withanangleof708relativetothewalkwaydirection)withadistanceof0.5m fromtheleftborderofthewalkway.Thefirstsensorwaspositionedat4mfromthe startofthewalkwayandtheothersensorswereplacedatinter-sensordistancesof 2.5m.Schematicsofthewalking-adaptabilitytasks:obstacleavoidancewithgait- dependent(B)andposition-dependentobstacles(C),suddenstops-and-starts(D) and goal-directed stepping with symmetric steppingstones (E), asymmetric steppingstones(F)andvariablesteppingstones(G).

targetsatdifferentpositionsonthewalkway,werealsoincludedto determinetheaveragedistancebetweenshoeedgesandtheankle forbothsystems.Thiscalibrationwasneededtodetermineseveral walking-adaptabilityoutcomemeasures(seebelow).

2.3. Datapre-processingandanalysis

Data pre-processing followed established procedures [7];

detailsabouttheprocedureandpre-processeddataarepresented asSupplementarymaterial.Duetoexcessivemissingdata,62out of2016trialswereexcludedfromfurtheranalysis,mainlyforthe gold-standardmotion-registrationsystem(i.e.,markerocclusion and/ororientationissues)andconcerningonesubject.

Thecontinuouswalking-adaptabilityoutcomemeasureswere available response time (ART) and margins of the trailing and leadinglimbduringobstaclecrossingfortheobstacle-avoidance task,ARTandmargintothestopcueforthesudden-stops-and- startstask,steplength,steppingaccuracyandwalkingspeedfor SSSandVSS,andleftandrightsteplengths,steppingaccuracyand walkingspeedforASS.Thesecontinuousoutcomemeasureswere calculatedfromspecificbodypoints’timeseries,estimatesoffoot contactandfootoffandsteplocations,asdetailedinTable1,for bothmeasurementsystemsalikeinanalignedcoordinatesystem, including the coordinates of obstacles, sudden-stop cues and targets.Forallcontinuousoutcomemeasures,statisticalanalyses were performed over averages over trials. For dichotomous outcomemeasures,steplocationswereextrapolatedtotheactual shoe dimensions based on the ankle-to-shoe calibration to determine whether or not obstacle-avoidance and sudden-stop trialsweresuccessfullyperformed,fromwhichsuccessrateswere deduced. Successful gait-dependent obstacle-avoidance maneu- verswereclassifiedasshort-strideorlong-stridestrategies[14].

2.4. Statisticalanalysis

Between-systems agreement wasdeterminedforcontinuous outcome measures using intraclass correlation coefficients for

absoluteagreement (ICC(A,1); [15]), withvalues above0.60 and 0.75 representing good and excellent agreement, respectively [16].Thisanalysisof between-systemsagreement wascomple- mentedbymeandifferencesandprecisionvaluesobtainedwitha Bland–Altmananalysis(i.e.,thebiasandthelimitsofagreement, respectively[17]).Fordichotomousoutcomemeasureswereport thepercentageofnon-matchedratings.

Sensitivity to task variation was examined using repeated- measuresANOVAsoncontinuousoutcomemeasuresofobstacle- avoidanceandgoal-directed-steppingtasks.ForARTandobstacle- avoidancemargins,aSystem(IWW,Optotrak)byObstacle(gait- dependent, position-dependent) by Limb (trailing, leading) re- peated-measuresANOVAwasconducted.Forsteplength,stepping accuracy and walkingspeed ofSSS, a Systemby Imposed step length(30,40,...,90)repeated-measuresANOVAwasconducted.

For left and right step lengths, stepping accuracy and walking speedofASS,aSystembyImposedstep-lengthasymmetry(L/R:

15/75,30/60,45/45,60/30,75/15)repeated-measuresANOVAwas conducted.Forsteplength,steppingaccuracyandwalkingspeedof VSS,aSystembyTrialrepeated-measuresANOVAwasconducted.

For the average stepping accuracy of the three goal-directed- stepping conditions, a System by Condition (SSS, ASS, VSS) repeated-measures ANOVA was conducted. One subject was excluded fromthe analysesof thegoal-directed-stepping tasks due to multiple trials with excessive missing values. The assumption of sphericity was checked according to Girden [18].IfGreenhouse–Geisser’sepsilonexceeded0.75,theHuynh–

Feldtcorrectionwasapplied;otherwisetheGreenhouse–Geisser correction was used. Main effects were examined with a LSD posthoctestforfactorswiththreelevelsandcontrastanalyses for factors with more than three levels.Paired-samples t-tests wereusedforsignificantinteractions.Effectsizeswerequantified withh²p.

Sensitivity to subject variation was examined by exploring speed-performancetrade-offs.WedeterminedPearson’scorrela- tions between self-selected walking speed and stepping accuracy for all goal-directed-stepping tasks and between the Table1

Calculationmethodsofcontinuouswalking-adaptabilityoutcomemeasures.

Outcomemeasure Unit Calculation

Obstacle-avoidancetask Availableresponsetime s Thedistanceofthenearestanteriorshoeedgetotheborderoftheobstacleatthe momentofitsappearancedividedbytheaveragewalkingspeedoverthesecondbefore itsappearance.

Obstacle-avoidancemargins cm Thedistanceoftheanteriorshoeedge(trailinglimb)andposteriorshoeedge(leading limb)ofthesteplocationstocorrespondingobstaclebordersduringobstaclecrossing.

Steplocationsweredeterminedasthemediananterior-posteriorpositionoftheankle jointduringthesingle-supportphase(i.e.,betweenfootoffandfootcontactofthe contralateralfoot)[7].Estimatesoffootcontactandfootoffweredefinedasthemaxima andminimaoftheanterior-posteriortimeseriesoftheanklesrelativetothatofthe spinebase[7,13].

Sudden-stops-and-startstask Availableresponsetime s Thedistanceofthenearestanteriorshoeedgetotheborderofthesudden-stopcueat themomentofitsappearancedividedbytheaveragewalkingspeedoverthesecond beforeitsappearance.

Sudden-stopmargin cm Theminimumdistanceoftheanteriorshoeedgetothecorrespondingsudden-stopcue borderduringtheperiodinwhichthecuewasvisible.

Goal-directed-steppingtask Steplength cm Themedianofthedifferencesintheanterior-posteriordirectionofconsecutivestep locations.Steplocationsweredeterminedasthemediananterior–posteriorpositionof theanklejointduringthesingle-supportphase(i.e.,betweenfootoffandfootcontactof thecontralateralfoot)[7].Estimatesoffootcontactandfootoffweredefinedasthe maximaandminimaoftheanterior–posteriortimeseriesoftheanklesrelativetothat ofthespinebase[7,13].

Steppingaccuracy cm Thestandarddeviationoverthesigneddeviationsbetweenthecenterofthefootand thecenterofthetargetatsteplocations,asdefinedinsteplength.Steppingaccuracy wasdeterminedoversteplocationsthatwereidentifiedforbothsystemstoensurea faircomparison.Thecenterofthefootwasdeterminedusingtheaveragedistance betweentheankleandthemiddleoftheshoe-size-matchedtargetsofthecalibration trials.

Walkingspeed cm/s Thedistancetraveledbetweenthestartandthe10-mlineofthewalkwaydividedby theduration,usingthedataofthespineshoulder.

speed-dependent ART and margins for obstacle-avoidance and sudden-stop tasks (i.e., significant positive correlations signal speed-performancetrade-offs).Wealsoassessedtheinfluenceof obstacle-avoidance and sudden-stop ratings on ART using a Systemby Rating (success, failure) repeated-measures ANOVA.

Inaddition,obstacle-avoidancesuccessrateswerecomparedwith aSystembyObstaclerepeated-measuresANOVA.

3. Results

3.1. Between-systemsagreement

Excellent between-systemsagreementwasobservedforART andmarginsforobstacle-avoidanceandsudden-stops-and-starts tasks,walkingspeedforallgoal-directed-steppingconditions(SSS, Table2

Agreementstatisticsforcontinuousoutcomemeasuresofobstacle-avoidance,sudden-stops-and-startsandgoal-directed-stepping(SSS,ASSandVSS)tasks.

InteractiveWalkwaymeanSD OptotraksystemmeanSD Bias[95%LoA] ICC(A,1)

Obstacle-avoidancetask

ART(s) Gait-dependent 0.7920.050 0.7770.049 0.015^*[0.0320.002] 0.945

Position-dependent 0.8340.075 0.8340.076 0.000[0.0230.024] 0.988

Margins(cm) Gait-dependent Trailinglimb 27.685.53 27.655.06 0.03[2.172.12] 0.980

Leadinglimb 11.685.45 12.785.26 1.11^*[1.353.56] 0.954

Position-dependent Trailinglimb 11.273.08 11.542.90 0.26[2.182.71] 0.913

Leadinglimb 8.974.91 9.824.87 0.85^*[1.393.09] 0.960

Sudden-stops-and-startstask

ART(s) 0.4970.067 0.4900.070 0.007^*[0.0350.021] 0.997

Margins(cm) 8.327.29 8.356.70 0.30[6.967.02] 0.876

SSS

Steplength(cm) 30 29.950.14 29.970.32 0.02[0.550.58] 0.339

40 39.960.18 40.000.28 0.04[0.610.68] 0.034

50 50.060.29 50.020.35 0.04[1.040.96] 0.276

60 60.020.38 59.890.48 0.13[1.210.95] 0.189

70 69.990.25 69.910.57 0.07[1.050.90] 0.376

80 79.890.28 79.760.48 0.13[1.100.84] 0.210

90 89.840.37 89.810.33 0.03[0.820.76] 0.367

Steppingaccuracy(cm) 30 1.770.41 1.870.38 0.10[0.550.75] 0.635

40 1.800.37 1.930.45 0.13[0.660.92] 0.503

50 1.810.37 2.000.47 0.20^*[0.490.88] 0.609

60 1.910.46 1.910.52 0.00[0.770.78] 0.686

70 1.910.41 1.990.49 0.08[0.640.80] 0.675

80 1.880.54 2.020.53 0.15[0.891.19] 0.498

90 2.020.55 2.120.56 0.10[0.590.78] 0.798

Walkingspeed(cm/s) 30 73.2312.95 72.8912.66 0.34^*[1.030.35] 0.999

40 86.9313.42 86.3713.04 0.57^*[1.480.35] 0.999

50 101.1414.11 100.4213.73 0.72^*[1.670.23] 0.998

60 112.2813.83 111.1913.28 1.09^*[2.570.39] 0.995

70 124.4013.38 123.2412.89 1.16^*[2.590.26] 0.995

80 136.7012.49 134.9712.07 1.73^*[3.000.46] 0.989

90 145.0712.07 143.4311.67 1.64^*[3.100.19] 0.989

ASS

Steplengthleft(cm) 15/75 21.383.66 19.753.92 1.63^*[4.301.03] 0.859

30/60 34.232.39 33.552.71 0.68[3.652.29] 0.803

45/45 44.721.17 44.501.76 0.22[3.032.59] 0.546

60/30 55.442.35 56.342.82 0.90^*[2.033.83] 0.793

75/15 67.442.96 69.883.58 2.45^*[0.965.86] 0.677

Steplengthright(cm) 15/75 68.573.84 70.163.96 1.60^*[1.414.61] 0.854

30/60 55.762.58 56.452.84 0.69[2.483.86] 0.803

45/45 45.371.24 45.391.87 0.01[2.852.88] 0.588

60/30 34.622.20 33.632.66 0.99^*[3.741.76] 0.777

75/15 22.802.89 19.963.56 2.83^*[6.370.71] 0.615

Steppingaccuracy(cm) 15/75 3.871.77 3.371.58 0.50^*[1.750.75] 0.891

30/60 2.871.13 2.651.08 0.21[1.541.11] 0.806

45/45 1.730.38 1.880.46 0.14[0.590.88] 0.584

60/30 3.021.03 2.791.03 0.23[1.200.74] 0.869

75/15 4.361.36 3.341.49 1.02^*[2.350.31] 0.709

Walkingspeed(cm/s) 15/75 90.8712.05 90.3311.81 0.54^*[1.340.25] 0.998

30/60 92.0113.61 91.4613.35 0.55^*[1.430.34] 0.999

45/45 91.7314.14 91.2013.96 0.53^*[1.340.28] 0.999

60/30 89.2314.18 88.7513.92 0.47^*[1.240.29] 0.999

75/15 87.8413.51 87.3113.25 0.53^*[1.330.26] 0.999

VSS

Steplength(cm) 45.540.82 45.490.85 0.05[0.960.86] 0.852

Steppingaccuracy(cm) 2.600.68 2.530.65 0.08[0.590.44] 0.920

Walkingspeed(cm/s) 97.8913.88 97.2513.56 0.64^*[1.510.23] 0.998

Meanvalues,between-subjectsstandarddeviations(SD)andagreementstatistics(bias,limitsofagreement[95%LoA]andintraclasscorrelationcoefficientforabsolute agreement[ICC(A,1)])forthecontinuousoutcomemeasuresoftheobstacle-avoidance,sudden-stops-and-startsandgoal-directed-steppingtasks.

ART=availableresponsetime,SSS=symmetricsteppingstones,ASS=asymmetricsteppingstones,VSS=variablesteppingstones.

* Significantbetween-systemsdifference(p<0.05).