Research ArticleComputer Vision-Based Approach for Quantifying OccupationalTherapistsrsquo Qualitative Evaluations of Postural Control

Hiromichi Hagihara 12 Naoto Ienaga 23 Daiki Enomoto4 Shuhei Takahata5

Hiroyuki Ishihara6 Haruka Noda 7 Koji Tsuda89 and Kei Terayama 91011

1Graduate School of Human and Environmental Studies Kyoto University Kyoto Japan2Research Fellow of Japan Society for the Promotion of Science Tokyo Japan3Graduate School of Science and Technology Keio University Yokohama Japan4LITALICO Inc Tokyo Japan5Department of Rehabilitation Hakuho College Nara Japan6Nippon Telegraph and Telephone Kanagawa Japan7Graduate School of Biomedical Sciences Nagasaki University Nagasaki Japan8Graduate School of Frontier Science University of Tokyo Tokyo Japan9RIKEN Center for Advanced Intelligence Project Tokyo Japan10Medical Sciences Innovation Hub Program RIKEN Cluster for Science Technology and Innovation Hub Kanagawa Japan11Graduate School of Medicine Kyoto University Kyoto Japan

Correspondence should be addressed to Hiromichi Hagihara hiromichihgmailcom and Kei Terayama keiterayamarikenjp

Received 22 August 2019 Accepted 3 March 2020 Published 27 April 2020

Academic Editor Claudia Hilton

Copyright copy 2020 Hiromichi Hagihara et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

This study aimed to leverage computer vision (CV) technology to develop a technique for quantifying postural control Aconventional quantitative index occupational therapistsrsquo qualitative clinical evaluations and CV-based quantitative indicesusing an image analysis algorithm were applied to evaluate the postural control of 34 typically developed preschoolers Theeffectiveness of the CV-based indices was investigated relative to current methods to explore the clinical applicability of theproposed method The capacity of the CV-based indices to reflect therapistsrsquo qualitative evaluations was confirmedFurthermore compared to the conventional quantitative index the CV-based indices provided more detailed quantitativeinformation with lower costs CV-based evaluations enable therapists to quantify details of motor performance that arecurrently observed qualitatively The development of such precise quantification methods will improve the science and practiceof occupational therapy and allow therapists to perform to their full potential

1 Introduction

The aim of occupational therapy is to promote peoplersquosparticipation in daily activities [1] This therapy is offeredto a wide range of clients from children to the elderly Occu-pational therapists conduct holistic assessments of clientsrsquocurrent body functions activities and participation in dailylives and draw mutual relationships among these aspects Inthe field of paediatric or school occupational therapy in

particular the relationships between body functions such aspostural control and everyday activities are very importantbecause body functions form the basis for the occupationalperformance of children with regard to skills such as self-care communication and subject learning [2] In factimpaired postural control has been widely reported in peoplediagnosed with developmental disorders such as autismspectrum disorder [3ndash5] attention-deficithyperactivitydisorder [6 7] and developmental coordination disorder

HindawiOccupational erapy InternationalVolume 2020 Article ID 8542191 9 pageshttpsdoiorg10115520208542191

[8 9] Therefore it is essential to devise evaluation indices ofbody functions that can enable therapists to grasp these diffi-culties comprehensively Such indices also help to establishevidence of the usefulness of occupational therapy in devel-oping outcome measures which are sensitive to occupationaltherapy interventions [10]

However current evaluation indices have serious limita-tions especially the quantitative ones First the granularityof conventional quantification methods is coarse In theclinical practice of paediatric occupational therapy oft-usedassessments such as the Movement Assessment Battery forChildren 2 [11] the Bruininks-Oseretsky Test of MotorProficiency 2 [12] or the Japanese Playful Assessment forNeuropsychological Abilities [13] generate quantitative out-comes that are simply the number of task successes durationtime or speed Although these indices can be useful they arenot sufficient to detail critical motor performances such asbalance antigravity posture or compensatory movementall of which therapists instead observe qualitatively Secondthe detailed qualitative evaluations that the occupationaltherapist would normally perform in the above assessmentsare often excluded from the indices themselves because suchevaluations are considered to be difficult to quantifyAlthough certain evaluation methods can quantify such qual-itative viewpoints using specialized equipment (eg bodysway meters) these methods are typically quite expensiveand have poor clinical availability [14] Recently some solu-tions have been proposed such as evaluations using con-sumer electronics devices [15 16] However occupationaltherapists still seem to adopt practical classification methodsbased on certain criteria for their clinical convenience insteadof quantifying the quality of motor performances Third cur-rent motor skill evaluation procedures require much timeand impose physical and mental burdens on patients result-ing in increased exhaustion and reduced motivation [17] Inparticular postural control basically needs the ability tomaintain postural balance as well as the ability to maintainpositions of body parts against the force of gravity to keepposture stable Therapists have utilized different tasks to eval-uate these abilities independently (eg one-foot standing forstatic postural balance and prone extension for antigravityposture respectively) [18 19] The development of tech-niques that enables therapists to observe these abilities simul-taneously in one task can reduce time costs and the burdensof patients

In this research we investigate the application ofcomputer vision (CV) technology to resolve these problemsRecent advances in deep learning technology have made sig-nificant progress in the field of monocular human postureestimation [20ndash23] These methods estimate the 2D or 3Dpositions of human key points (eg joints) from only RGBimages or movies CV studies capitalizing on such pose esti-mation methods enabled researchers to easily quantifyhuman postures or bodymovements [24ndash26] In the past sev-eral years CV techniques have been applied to medical fieldswhich include for example diagnosis of autism spectrumdisorder [27 28] or detection of gait abnormalities [29]Although detailed quantitative analyses of postures and bodymovements have conventionally required special equipment

such as motion captures or force plates these technologicalinnovations have the potential to decrease their expenseand ease their implementation by therapists in the clinicalsetting

The purpose of this research was to develop a techniquefor quantifying postural control by using a pose estimationalgorithm called OpenPose [20] and to explore the proposedtechniquersquos clinical applicability by comparing it to individ-ual occupational therapistsrsquo qualitative clinical evaluationsAs the first step toward this goal we conducted a simple pos-tural control evaluation a static posture maintenance taskfor typically developing children

This research contributes to the occupational therapyfield first by demonstrating the possibility of more accurateand detailed quantitative evaluations and second by simpli-fying conventional evaluation tasks with the elimination ofspecial equipment and reduced time investments imposingless burden on clients Furthermore this research may accel-erate collaborations between occupational therapy and com-puter vision fields

2 Materials and Methods

21 Design Typically developed children performed adevelopmental task standardised in Japan to evaluate pos-tural stability and three types of indices were calculated (1)a conventional quantitative index measuring duration time(DT) (2) therapistsrsquo qualitative clinical evaluations (TQCE)using the 7-point Likert scale and (3) CV-based quantitativeindices based on image analysis algorithms for static posturalbalance (SPB) and antigravity (AG) We then investigatedhow closely the CV-based scores reflected the TQCE (analy-sis 1) and how effective the CV-based model was comparedto the DT-based model (analysis 2 Figure 1) As the collabo-rative and interdisciplinary research both occupational ther-apists and CV engineers participated in every step of theresearch flow in order to avoid mutual lack of understandingdue to the explicit division of work

22 Participants Thirty-four preschoolers (14 girls meanage = 47 SD = 10 range = 3 ndash 6 years old) from two nurseryschools in Japan participated in this research An additionalfive children also participated but were excluded from thefinal sample because of their difficulty maintaining the taskposture without assistance All participants in this researchwere considered to be developing typically Each participantrsquosparents provided written informed consent and this researchwas approved by the Research Ethics Committee of HakuhoCollege (18012)

23 Instrumentation The Japanese Playful Assessment forNeuropsychological Abilities (JPAN) is a developmentalassessment battery for evaluating childrenrsquos sensory integra-tion abilities [13 30 31] which was developed in Japan basedon previous tests developed in the United States such as theSensory Integration and Praxis Tests [32] and standardisedfrom a large sample (489 Japanese children of 4ndash10 yearsold) The JPAN consists of 32 subtests (six for equilibriumand antigravity posture seven for somatosensory 15 for

2 Occupational Therapy International

praxis and four for eye-hand coordination and visual per-ception) and these have been confirmed for high interraterreliability (intraclass correlations in the range of 081ndash100for each subtest) and internal consistency (Cronbachrsquos alphaof 075 for all subtests) [33]

Among JPAN subtests we selected the One Arm andOne Leg Balance as the experimental task for this researchin the domain of equilibrium and antigravity posture Thistask required a participant to maintain what is called ldquobirddog posturerdquo for up to 60 s on each side in which heshelifted the arm on one side and the leg on the other side froma four-point crawling posture (see Figure 1) This taskrequired an unfamiliar balance [34] and was therefore con-sidered to remove the influence of participantsrsquo splinter skillsto some extent Despite the need for multiple functions tomaintain this posture including static postural balance andantigravity the task has been quantitatively assessed onlyby DT which was the primary reason for selecting this task

24 Task Procedure Each participant worked on the taskindividually in a vacant room in a nursery school performingthe task twice (once on each side) according to the JPANmanual (for up to 60 s on each side) The examiner cheeredparticipants up to encourage them to maintain the bird dogposture as long as they could (eg ldquoLift your armrdquo) Taskperformances were videotaped from the side and at approxi-mately the same height on each participantrsquos trunk so thatvideo frames included each participantrsquos whole body on thesagittal plane

25 Applied Indices Three types of indices were calculated toexplore which indices most appropriately reflected therapistsrsquoqualitative clinical evaluations

251 DT The DT for which each participant maintained theposture was measured by detecting the frames in which par-

ticipants fully extended their arms and legs and those inwhich any part of these body parts was attached to the floorIf 60 s elapsed without putting any part of these body parts onthe floor the task was aborted and the duration time wasrecorded as 60 s This index is a conventional quantitativeoutcome used in the JPAN

252 TQCE Three paediatric occupational therapists(Hagihara Noda and Takahata authors of this article theirclinical experiences were approximately 4 2 and 11respectively) evaluated the quality of the task performancebased on the clinical viewpoint of each therapist followingthe 7-point Likert scale (where a score of 7 indicates themost superior performance) These therapists watchedand scored the recorded videos individually so that theydid not know other therapistsrsquo scoring In addition theywatched the raw recorded videos that had not been analyzedby OpenPose yet Hence they did not know the outcomes ofCV-based quantitative indices for each participant at all whenevaluating A high interrater reliability was confirmed(ICC eth2 1THORN = 0892) The three therapistsrsquo evaluations wereaveraged to determine the TQCE index

253 CV-Based Approach First three CV engineers (IenagaTerayama and Ishihara authors of this article) conductedthe key point detection of body parts with OpenPose version140 OpenPose is an open-source library for real-time mul-tiperson key point detection of a body face and hands usinga convolutional neural network [20] Using a method calledPart Affinity Fields (PAFs) OpenPose can take into accountthe positional relationship of each key point that is thehuman skeletal structure This algorithm contributes toimproving the accurate estimation of key-point positionsOpenPose can estimate the 2D positions of 25 body keypoints which roughly correspond to joints (eg key point 6is left elbow see Figure 1) The confidence value of each

Captured video

Pose estimationby OpenPose

(i) 25 key points(ii) Confidence value

Detected keypoints

Static postural balance (SPB) score Antigravity score (AG)

Averaging

Multipleregression

analysis

Time

Analysis 1

Analysis 2

Durationtime (DT)

Therapistsrsquo qualitativeclinical evaluation

(TQCE)

120593

997903120579997903 + 997903120593997903

120579Correlation between TQCEand CV-based evaluation

Effectiveness ofCV-based scoring

Figure 1 Research flowchart

3Occupational Therapy International

key point for each frame is automatically calculated rangingfrom 0 to 1 (a higher value indicates higher confidence) sothat the degree of precision of key-point estimation can becomprehensible For example in Figure 1 all the key pointsused for subsequent analysis were detected with high confi-dence values (gt05)

OpenPose can detect the key points more accuratelywhen the head of a person is above and the legs below dueto the effect of the training data The input video was rotated90deg to turn the participantsrsquo heads upward since the partici-pants kept the bird dog posture as shown in Figure 1 Thekey points were then rotated 90deg back after being detectedby OpenPose To eliminate the detection noise key pointswhose confidence values were less than 05 were removedin each frame and interpolated linearly

Because occupational therapists evaluate SPB and AG inaddition to duration time we defined these items as shownbelow

254 Static Postural Balance (SPB) We first calculate themoving distance Dprime

D = 1n minus 1

nminus1

i=1

diiminus1bl

eth1THORN

Dprime = Dw +De +Da +Dketh THORN4 eth2THORN

D is the moving distance of a key point averaged in theframe direction (n represents the number of frames of thevideo) The moving distance diiminus1 between two consecutiveframes i i minus 1 is divided by the body length bl The bodylength is the average value of the distance between two keypoints corresponding to the neck and hip in all frames ofthe video The reason for dividing the moving distance bythe body length is to reduce the influences of the differencesin the heights of the subjects and in the distances between thecamera and the subjects There are actually four Ds for thewrist key point Dw and the elbow key point De of theextended arm and the ankle key point Da and the knee keypoint Dk of the extended leg SPB is the standardised value(z-score) obtained as follows

SPB =log 1Dprime

minus μ

σ eth3THORN

where μ and σ indicate the average and standard deviation oflog eth1DprimeTHORN respectively The average and standard deviationof SPB are 0 and 1 respectively because SPB is standardisedto eliminate the influence of numerical magnitude for multi-ple regression analysis with other indices High SPB indicateshigh static postural stability

255 Antigravity Posture (AG) AG was calculated as

a = minus1nnminus1

i=0φj j + θj jeth THORN eth4THORN

where jφj and jθj are the absolute angles between the floorand the extended arm and leg respectively as shown inFigure 1 Here ldquoarmrdquo refers to the vector connecting the neckposition to the average position of the wrist and elbow andldquolegrdquo refers to the vector connecting the hip position to theaverage position of the ankle and knee Because these areabsolute values the angle increases whether the limb is raisedor lowered a is the negative of the sum of the two anglesaveraged over the frame direction and AG is the standar-dised value (z-score) of a High AG indicates high antigravityposture

26 Data Analysis First we investigated how closely theCV-based scores (SPB and AG) reflected TQCE usingSpearmanrsquos rank correlation coefficient (analysis 1) Nextwe compared the effectiveness of the CV-based model tothat of the DT-based model (analysis 2) SPB AG andDT together or DT alone were used as the explanatoryvariables and regression analyses were performed usingTQCE as the objective variable with subsequent correla-tion analysis between the predicted value of the regressionanalysis and TQCE assuming that the test time wasunchanged

3 Results and Discussion

31 Degree to which CV-Based Scores Reflected TQCE Basedon the analyses using Spearmanrsquos rank correlation coeffi-cient SPB showed a significant positive correlation withTQCE (rs = 0654 p lt 0001) AG also showed a significantthough weak positive correlation with TQCE (rs = 0382p = 00013) TQCE tended to be higher when both of theseCV-based scores were high (Figure 2)

32 Multiple Regression Model Reflecting TQCE To comparethe evaluations using only the conventional quantitative index(DT) with the evaluations using all three indices (DT and theCV-based scores) we set TQCE as the objective variable andcalculated the model using DT as the explanatory variable ina single regression analysis in comparison with the modelusing SPB AG and DT as the explanatory variables in a mul-tiple regression analysis DT alone had a significant positive

ndash2

ndash2

ndash1

ndash1 0

0

1

AG

2

1 21

2

3

TQCE

4

5

SPB

Figure 2 Scatter plot of SPB and AG with colour bar of TQCE

4 Occupational Therapy International

correlation with TQCE (rs = 0886 p lt 0001) However thevalue calculated by multiple regression analysis whichincluded the CV-based scores showed a stronger significantpositive correlation with TQCE (rs = 0914 p lt 0001)

For test times shorter than 60 s the difference betweenthe CV- and DT-based scores was even more pronounced(Figure 3(a)) When the test time was reduced to 20 s theCV-based score showed a higher positive correlation

090

085

080

075

070

065

10 20 30 40Test time (sec)

CV-based scoreDT

50 60Sp

earm

anrsquos

rank

corr

elat

ion

coeffi

cien

t

(b)

(d)

(c)

(e)

(a)

CV-based score

rs = 0878

ndash2 ndash1 0

1

2

3TQCE

4

5

1

(b)

1

2

3TQCE

4

5

CV-based scorendash2 ndash1 0 1

rs = 0914

(c)

DT

1

5 10 15 20

2

3TQCE

4

5rs = 0789

(d)

DT0 20 40 60

1

2

3TQCE

4

5rs = 0886

(e)

Figure 3 Comparison of the CV- and DT-based models (a) Correlation coefficient between TQCE and CV- (blue) or DT-based (orange)models for every 10 s of test time (b) Scatter plot of CV-based scores and TQCE at the test time of 20 s (c) Scatter plot of CV-basedscores and TQCE at the test time of 60 s (d) Scatter plot of DT-based scores and TQCE at the test time of 20 s (e) Scatter plot of DT-based scores and TQCE at the test time of 60 s lowastlowastlowastp lt 0001

5Occupational Therapy International

(rs = 0878 p lt 0001 see Figures 3(b) and 3(c)) than DTalone (rs = 0789 p lt 0001 see Figures 3(d) and 3(e)) Inall 68 trials (34 participants times two sides) the number of trialsfor which DT was equal or more than 60 s was 18 (26) andthe number for which DT was equal or more than 20 swas 37 (54) The results of each regression analysis areshown in Table 1

33 Effectiveness of CV-Based Indices The results of thisresearch revealed the following two points First both SPBand AG had a significant positive correlation with TQCEand TQCE tended to be higher when both of these CV-based scores were high indicating that these indices quanti-fied TQCE well Second the multiple regression analysisand subsequent correlation analysis showed that the CV-based model reflected the TQCE to a greater degree thanthe conventional DT-based model indicating that evalua-tions using CV-based scores provide more detailed quantita-tive information These results showed the possibility ofovercoming the severe problems with current quantitativeevaluation indices which are useful for screening but arenot sufficient for verifying clinical effects or depicting detailsof motor performances including postural control Althougha certain proportion of children with developmental disor-ders have difficulties with postural control [3ndash9] they oftendemonstrate the maximum performance in terms of currentquantitative scores resulting in a ceiling effect [17] and theirdifficulties with postural control can only be identified bytherapistsrsquo clinical observations By using CV-based scorestherapists can quantitatively evaluate their postural controleven when they achieve maximumDT Although the correla-tion coefficient between AG and TQCE was significant butlow this result may have occurred because therapists evalu-ated AG posture only after confirming the achievement ofinitial evaluation items such as durability or static posturalbalance

The developed CV-based scores SPB and AG will con-tribute not only to identifying essential factors of posturalcontrol that have negative effects on childrenrsquos everydayactivities but also to establishing more valid evidence forthe efficacy of occupational therapy Body functions such aspostural control are the basis for childrenrsquos occupational per-formance on skills such as self-care communication andsubject learning [2] Furthermore children with develop-mental disorders especially developmental coordination dis-order often suffer from psychological problems including

lower self-esteem [35 36] anxiety and depression [37 38]The qualitative improvement of postural control might thuscontribute to improving their activities and participationand such improvements might be grasped by CV-basedscores because these detailed quantitative indices reflect thequalitative aspects of postural control These scores will alsomeet the need for developing functional outcome measureswhich are sensitive to occupational therapy interventions[10] Studies on the postural control of people with develop-mental disorders have focused on a specific posture onlysuch as standing on one leg (eg [39]) or on both legs (eg[40 41]) the results of which cannot be widely applied tovarious postures In contrast the CV-based approach willbroaden opportunities to advance detailed and quantitativeevaluations of the qualitative aspects of a broad range of pos-tures important for early postural development and activitiesin daily living

34 Applicability to the Development of Lower Cost and moreDetailed Assessments The degree to which the CV-basedmodel reflected TQCE was hardly reduced even when the testtime was shortened Furthermore the precision with whichthe models reflected TQCE were almost equal between theDT-based model at the test time of 60 s and the CV-basedmodel at the test time of 20 s This suggests the possibilityof making the current assessment task procedure easier andsimpler to reduce the time cost of testing and the physicaland psychological burden on clients In fact the high costand burden of conventional standardised assessments forchildren with poor motor skills are regarded as a clinicalproblem because longer test times cause fatigue and loss ofmotivation [17] The CV-based approach shortens the testtime and also yields more detailed quantitative informationthat is useful for both scientific research and clinical practice

The experimental task ldquoOne Arm and One Leg Balancerdquoused in this research simultaneously evaluates multipleaspects of postural control including at least SPB AG andDT Although developing several tasks to evaluate eachaspect separately may be desirable in particularly complexcases this increases the cost and physical and psychologicalburdens on patients and thus has poor clinical applicabilityThe research presented herein demonstrates that the pro-posed indices resolve and quantify multiple aspects of pos-tural control for multiple important aspects simultaneouslyoffering substantial reductions in the cost of the evaluationwithout compromising the amount of information

Table 1 Regression analysis results

Model using SPB AG and DT Model using only DTTest time 20 s Test time 60 s Test time 20 s Test time 60 s

Variables β (SE) β (SE) β (SE) β (SE)

SPB 035 (0083)lowastlowastlowast 025 (0059)lowastlowastlowast

AG 025 (0070)lowastlowastlowast 022 (0050)lowastlowastlowast

DT 050 (0068)lowastlowastlowast 069 (0060)lowastlowastlowast 076 (0081)lowastlowastlowast 088 (0059)lowastlowastlowast

Adjusted R2 069lowastlowastlowast 084lowastlowastlowast 057lowastlowastlowast 077lowastlowastlowast

lowastlowastlowastp lt 0001 therapistsrsquo qualitative clinical evaluation (TQCE) was set as the objective variable SPB static postural stability AG antigravity posture DTduration time SE standard error

6 Occupational Therapy International

35 Limitations and Future Work This research has severallimitations because we conducted a simple task on typicallydeveloping children so that the clinical applicability of CV-based quantification and analysis could be easily exploredA larger and randomized sample of children with and with-out developmental disorders must be beneficial to confirmthe validity of this CV-based approach Furthermore therelationship between the seriousness of motor skill impair-ment or its developmental change and CV-based indicesshould be investigated for diagnostic use Whether the indi-ces for postural control developed in this research can beapplied to other tasks or postures must also be confirmedApplication software that facilitates therapistsrsquo quantitativeevaluations or verifications of the effectiveness of interven-tions would be especially important clinically Despite theselimitations this research advances the fusion of occupationalscience and practice by bringing together the fields of occu-pational therapy and CV In future works we are planningto focus on the clinical usability of the CV-based approachproposed in this research (eg user interface) Furthermorewe intend to develop a versatile quantification approach thatcan be used not only in standardised assessments but also inactual occupational activities and is applicable to variousdisorders The development of more precise quantificationmethods provides a basis for considering more effectiveinterventions and contributes to realizing occupational ther-apistsrsquo full potential

4 Conclusions

This study confirmed the close match between the results ofCV-based indices (SPB and AG) and TQCE Furthermorecompared with assessments using only the conventionalquantitative index (DT) evaluations that included CV-based indices provided more detailed quantitative informa-tion with lower time costs This research showed the possibil-ity of overcoming the problems associated with currentquantitative evaluations that they are insufficient for verify-ing clinical effects or detailing motor performances whichtherapists must observe qualitatively The results of this studyhave the following implications for occupational therapypractice

(i) Application of CV-based indices provides detailedquantitative and shareable information useful fornot only screening assessments but also for verifyinginterventions more precisely

(ii) CV-based approaches can reduce the time cost andthe physical and psychological burdens on clientsimposed by the performance of the assessment task

(iii) Leveraging CV technology enables the direct quanti-tative evaluation of motor performance duringactual occupational activities of daily living

This research also showed the possibility of making con-ventional evaluation tasks easier and simpler without specialequipment so that therapists can reduce time costs and thephysical and psychological burdens on clients This integra-

tion of occupational therapy and CV advances the scienceand practice of occupational therapy so that therapists canperform to their full potential

Data Availability

The data used to support the findings of this study are Excelfiles which consist of DT CV-based indices (SPB and AG)and TQCE of each participant Please contact the authorsof the article to address the data

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Authorsrsquo Contributions

H Hagihara and N Ienaga contributed equally to this work

Acknowledgments

We would like to thank T Kato and M Sakagami for theirinsightful comments and we are also grateful to our studyparticipants We used RAIDEN a supercomputer system ofAIP (RIKEN) to conduct pose estimation and analysis Thisresearch was supported by a Grant-in-Aid from the JapanSociety for the Promotion of Science (JSPS) Research Fellows(18J21948 and 17J05489)

References

[1] World Federation of Occupational Therapists Definitions ofoccupational therapy from member organisations WorldFederation of Occupational Therapists United Kingdom2019 July 2019 httpswwwwfotorgresourcesdefinitions-of-occupational-therapy-from-member-organisations

[2] A J Ayres Sensory Integration and the Child WesternPsychological Services Los Angeles (CA) 1979

[3] D Green T Charman A Pickles et al ldquoImpairment in move-ment skills of children with autistic spectrum disordersrdquoDevelopmental Medicine and Child Neurology vol 51 no 4pp 311ndash316 2009

[4] R Iwanaga C Kawasaki and R Tsuchida ldquoBrief report com-parison of sensory-motor and cognitive function betweenautism and Asperger syndrome in preschool childrenrdquo Journalof Autism and Developmental Disorders vol 30 no 2 pp 169ndash174 2000

[5] C P Whyatt and C M Craig ldquoMotor skills in children aged7ndash10 years diagnosed with autism spectrum disorderrdquo Journalof Autism and Developmental Disorders vol 42 no 9pp 1799ndash1809 2012

[6] J Konicarova P Bob and J Raboch ldquoBalance deficits andADHD symptoms in medication-naiumlve school-aged boysrdquoNeuropsychiatric Disease and Treatment vol 10 pp 85ndash882014

[7] M P Bucci C Stordeur E Acquaviva H Peyre andR Delorme ldquoPostural instability in children with ADHD isimproved by methylphenidaterdquo Frontiers in Neurosciencevol 10 p 163 2016

7Occupational Therapy International

praxis and four for eye-hand coordination and visual per-ception) and these have been confirmed for high interraterreliability (intraclass correlations in the range of 081ndash100for each subtest) and internal consistency (Cronbachrsquos alphaof 075 for all subtests) [33]

Among JPAN subtests we selected the One Arm andOne Leg Balance as the experimental task for this researchin the domain of equilibrium and antigravity posture Thistask required a participant to maintain what is called ldquobirddog posturerdquo for up to 60 s on each side in which heshelifted the arm on one side and the leg on the other side froma four-point crawling posture (see Figure 1) This taskrequired an unfamiliar balance [34] and was therefore con-sidered to remove the influence of participantsrsquo splinter skillsto some extent Despite the need for multiple functions tomaintain this posture including static postural balance andantigravity the task has been quantitatively assessed onlyby DT which was the primary reason for selecting this task

24 Task Procedure Each participant worked on the taskindividually in a vacant room in a nursery school performingthe task twice (once on each side) according to the JPANmanual (for up to 60 s on each side) The examiner cheeredparticipants up to encourage them to maintain the bird dogposture as long as they could (eg ldquoLift your armrdquo) Taskperformances were videotaped from the side and at approxi-mately the same height on each participantrsquos trunk so thatvideo frames included each participantrsquos whole body on thesagittal plane

25 Applied Indices Three types of indices were calculated toexplore which indices most appropriately reflected therapistsrsquoqualitative clinical evaluations

251 DT The DT for which each participant maintained theposture was measured by detecting the frames in which par-

ticipants fully extended their arms and legs and those inwhich any part of these body parts was attached to the floorIf 60 s elapsed without putting any part of these body parts onthe floor the task was aborted and the duration time wasrecorded as 60 s This index is a conventional quantitativeoutcome used in the JPAN

252 TQCE Three paediatric occupational therapists(Hagihara Noda and Takahata authors of this article theirclinical experiences were approximately 4 2 and 11respectively) evaluated the quality of the task performancebased on the clinical viewpoint of each therapist followingthe 7-point Likert scale (where a score of 7 indicates themost superior performance) These therapists watchedand scored the recorded videos individually so that theydid not know other therapistsrsquo scoring In addition theywatched the raw recorded videos that had not been analyzedby OpenPose yet Hence they did not know the outcomes ofCV-based quantitative indices for each participant at all whenevaluating A high interrater reliability was confirmed(ICC eth2 1THORN = 0892) The three therapistsrsquo evaluations wereaveraged to determine the TQCE index

253 CV-Based Approach First three CV engineers (IenagaTerayama and Ishihara authors of this article) conductedthe key point detection of body parts with OpenPose version140 OpenPose is an open-source library for real-time mul-tiperson key point detection of a body face and hands usinga convolutional neural network [20] Using a method calledPart Affinity Fields (PAFs) OpenPose can take into accountthe positional relationship of each key point that is thehuman skeletal structure This algorithm contributes toimproving the accurate estimation of key-point positionsOpenPose can estimate the 2D positions of 25 body keypoints which roughly correspond to joints (eg key point 6is left elbow see Figure 1) The confidence value of each

Captured video

Pose estimationby OpenPose

(i) 25 key points(ii) Confidence value

Detected keypoints

Static postural balance (SPB) score Antigravity score (AG)

Averaging

Multipleregression

analysis

Time

Analysis 1

Analysis 2

Durationtime (DT)

Therapistsrsquo qualitativeclinical evaluation

(TQCE)

120593

997903120579997903 + 997903120593997903

120579Correlation between TQCEand CV-based evaluation

Effectiveness ofCV-based scoring

Figure 1 Research flowchart

3Occupational Therapy International

key point for each frame is automatically calculated rangingfrom 0 to 1 (a higher value indicates higher confidence) sothat the degree of precision of key-point estimation can becomprehensible For example in Figure 1 all the key pointsused for subsequent analysis were detected with high confi-dence values (gt05)

OpenPose can detect the key points more accuratelywhen the head of a person is above and the legs below dueto the effect of the training data The input video was rotated90deg to turn the participantsrsquo heads upward since the partici-pants kept the bird dog posture as shown in Figure 1 Thekey points were then rotated 90deg back after being detectedby OpenPose To eliminate the detection noise key pointswhose confidence values were less than 05 were removedin each frame and interpolated linearly

Because occupational therapists evaluate SPB and AG inaddition to duration time we defined these items as shownbelow

254 Static Postural Balance (SPB) We first calculate themoving distance Dprime

D = 1n minus 1

nminus1

i=1

diiminus1bl

eth1THORN

Dprime = Dw +De +Da +Dketh THORN4 eth2THORN

D is the moving distance of a key point averaged in theframe direction (n represents the number of frames of thevideo) The moving distance diiminus1 between two consecutiveframes i i minus 1 is divided by the body length bl The bodylength is the average value of the distance between two keypoints corresponding to the neck and hip in all frames ofthe video The reason for dividing the moving distance bythe body length is to reduce the influences of the differencesin the heights of the subjects and in the distances between thecamera and the subjects There are actually four Ds for thewrist key point Dw and the elbow key point De of theextended arm and the ankle key point Da and the knee keypoint Dk of the extended leg SPB is the standardised value(z-score) obtained as follows

SPB =log 1Dprime

minus μ

σ eth3THORN

where μ and σ indicate the average and standard deviation oflog eth1DprimeTHORN respectively The average and standard deviationof SPB are 0 and 1 respectively because SPB is standardisedto eliminate the influence of numerical magnitude for multi-ple regression analysis with other indices High SPB indicateshigh static postural stability

255 Antigravity Posture (AG) AG was calculated as

a = minus1nnminus1

i=0φj j + θj jeth THORN eth4THORN

where jφj and jθj are the absolute angles between the floorand the extended arm and leg respectively as shown inFigure 1 Here ldquoarmrdquo refers to the vector connecting the neckposition to the average position of the wrist and elbow andldquolegrdquo refers to the vector connecting the hip position to theaverage position of the ankle and knee Because these areabsolute values the angle increases whether the limb is raisedor lowered a is the negative of the sum of the two anglesaveraged over the frame direction and AG is the standar-dised value (z-score) of a High AG indicates high antigravityposture

26 Data Analysis First we investigated how closely theCV-based scores (SPB and AG) reflected TQCE usingSpearmanrsquos rank correlation coefficient (analysis 1) Nextwe compared the effectiveness of the CV-based model tothat of the DT-based model (analysis 2) SPB AG andDT together or DT alone were used as the explanatoryvariables and regression analyses were performed usingTQCE as the objective variable with subsequent correla-tion analysis between the predicted value of the regressionanalysis and TQCE assuming that the test time wasunchanged

3 Results and Discussion

31 Degree to which CV-Based Scores Reflected TQCE Basedon the analyses using Spearmanrsquos rank correlation coeffi-cient SPB showed a significant positive correlation withTQCE (rs = 0654 p lt 0001) AG also showed a significantthough weak positive correlation with TQCE (rs = 0382p = 00013) TQCE tended to be higher when both of theseCV-based scores were high (Figure 2)

32 Multiple Regression Model Reflecting TQCE To comparethe evaluations using only the conventional quantitative index(DT) with the evaluations using all three indices (DT and theCV-based scores) we set TQCE as the objective variable andcalculated the model using DT as the explanatory variable ina single regression analysis in comparison with the modelusing SPB AG and DT as the explanatory variables in a mul-tiple regression analysis DT alone had a significant positive

ndash2

ndash2

ndash1

ndash1 0

0

1

AG

2

1 21

2

3

TQCE

4

5

SPB

Figure 2 Scatter plot of SPB and AG with colour bar of TQCE

4 Occupational Therapy International

correlation with TQCE (rs = 0886 p lt 0001) However thevalue calculated by multiple regression analysis whichincluded the CV-based scores showed a stronger significantpositive correlation with TQCE (rs = 0914 p lt 0001)

For test times shorter than 60 s the difference betweenthe CV- and DT-based scores was even more pronounced(Figure 3(a)) When the test time was reduced to 20 s theCV-based score showed a higher positive correlation

090

085

080

075

070

065

10 20 30 40Test time (sec)

CV-based scoreDT

50 60Sp

earm

anrsquos

rank

corr

elat

ion

coeffi

cien

t

(b)

(d)

(c)

(e)

(a)

CV-based score

rs = 0878

ndash2 ndash1 0

1

2

3TQCE

4

5

1

(b)

1

2

3TQCE

4

5

CV-based scorendash2 ndash1 0 1

rs = 0914

(c)

DT

1

5 10 15 20

2

3TQCE

4

5rs = 0789

(d)

DT0 20 40 60

1

2

3TQCE

4

5rs = 0886

(e)

Figure 3 Comparison of the CV- and DT-based models (a) Correlation coefficient between TQCE and CV- (blue) or DT-based (orange)models for every 10 s of test time (b) Scatter plot of CV-based scores and TQCE at the test time of 20 s (c) Scatter plot of CV-basedscores and TQCE at the test time of 60 s (d) Scatter plot of DT-based scores and TQCE at the test time of 20 s (e) Scatter plot of DT-based scores and TQCE at the test time of 60 s lowastlowastlowastp lt 0001

5Occupational Therapy International

(rs = 0878 p lt 0001 see Figures 3(b) and 3(c)) than DTalone (rs = 0789 p lt 0001 see Figures 3(d) and 3(e)) Inall 68 trials (34 participants times two sides) the number of trialsfor which DT was equal or more than 60 s was 18 (26) andthe number for which DT was equal or more than 20 swas 37 (54) The results of each regression analysis areshown in Table 1

33 Effectiveness of CV-Based Indices The results of thisresearch revealed the following two points First both SPBand AG had a significant positive correlation with TQCEand TQCE tended to be higher when both of these CV-based scores were high indicating that these indices quanti-fied TQCE well Second the multiple regression analysisand subsequent correlation analysis showed that the CV-based model reflected the TQCE to a greater degree thanthe conventional DT-based model indicating that evalua-tions using CV-based scores provide more detailed quantita-tive information These results showed the possibility ofovercoming the severe problems with current quantitativeevaluation indices which are useful for screening but arenot sufficient for verifying clinical effects or depicting detailsof motor performances including postural control Althougha certain proportion of children with developmental disor-ders have difficulties with postural control [3ndash9] they oftendemonstrate the maximum performance in terms of currentquantitative scores resulting in a ceiling effect [17] and theirdifficulties with postural control can only be identified bytherapistsrsquo clinical observations By using CV-based scorestherapists can quantitatively evaluate their postural controleven when they achieve maximumDT Although the correla-tion coefficient between AG and TQCE was significant butlow this result may have occurred because therapists evalu-ated AG posture only after confirming the achievement ofinitial evaluation items such as durability or static posturalbalance

The developed CV-based scores SPB and AG will con-tribute not only to identifying essential factors of posturalcontrol that have negative effects on childrenrsquos everydayactivities but also to establishing more valid evidence forthe efficacy of occupational therapy Body functions such aspostural control are the basis for childrenrsquos occupational per-formance on skills such as self-care communication andsubject learning [2] Furthermore children with develop-mental disorders especially developmental coordination dis-order often suffer from psychological problems including

lower self-esteem [35 36] anxiety and depression [37 38]The qualitative improvement of postural control might thuscontribute to improving their activities and participationand such improvements might be grasped by CV-basedscores because these detailed quantitative indices reflect thequalitative aspects of postural control These scores will alsomeet the need for developing functional outcome measureswhich are sensitive to occupational therapy interventions[10] Studies on the postural control of people with develop-mental disorders have focused on a specific posture onlysuch as standing on one leg (eg [39]) or on both legs (eg[40 41]) the results of which cannot be widely applied tovarious postures In contrast the CV-based approach willbroaden opportunities to advance detailed and quantitativeevaluations of the qualitative aspects of a broad range of pos-tures important for early postural development and activitiesin daily living

34 Applicability to the Development of Lower Cost and moreDetailed Assessments The degree to which the CV-basedmodel reflected TQCE was hardly reduced even when the testtime was shortened Furthermore the precision with whichthe models reflected TQCE were almost equal between theDT-based model at the test time of 60 s and the CV-basedmodel at the test time of 20 s This suggests the possibilityof making the current assessment task procedure easier andsimpler to reduce the time cost of testing and the physicaland psychological burden on clients In fact the high costand burden of conventional standardised assessments forchildren with poor motor skills are regarded as a clinicalproblem because longer test times cause fatigue and loss ofmotivation [17] The CV-based approach shortens the testtime and also yields more detailed quantitative informationthat is useful for both scientific research and clinical practice

The experimental task ldquoOne Arm and One Leg Balancerdquoused in this research simultaneously evaluates multipleaspects of postural control including at least SPB AG andDT Although developing several tasks to evaluate eachaspect separately may be desirable in particularly complexcases this increases the cost and physical and psychologicalburdens on patients and thus has poor clinical applicabilityThe research presented herein demonstrates that the pro-posed indices resolve and quantify multiple aspects of pos-tural control for multiple important aspects simultaneouslyoffering substantial reductions in the cost of the evaluationwithout compromising the amount of information

Table 1 Regression analysis results

Model using SPB AG and DT Model using only DTTest time 20 s Test time 60 s Test time 20 s Test time 60 s

Variables β (SE) β (SE) β (SE) β (SE)

SPB 035 (0083)lowastlowastlowast 025 (0059)lowastlowastlowast

AG 025 (0070)lowastlowastlowast 022 (0050)lowastlowastlowast

DT 050 (0068)lowastlowastlowast 069 (0060)lowastlowastlowast 076 (0081)lowastlowastlowast 088 (0059)lowastlowastlowast

Adjusted R2 069lowastlowastlowast 084lowastlowastlowast 057lowastlowastlowast 077lowastlowastlowast

lowastlowastlowastp lt 0001 therapistsrsquo qualitative clinical evaluation (TQCE) was set as the objective variable SPB static postural stability AG antigravity posture DTduration time SE standard error

6 Occupational Therapy International

35 Limitations and Future Work This research has severallimitations because we conducted a simple task on typicallydeveloping children so that the clinical applicability of CV-based quantification and analysis could be easily exploredA larger and randomized sample of children with and with-out developmental disorders must be beneficial to confirmthe validity of this CV-based approach Furthermore therelationship between the seriousness of motor skill impair-ment or its developmental change and CV-based indicesshould be investigated for diagnostic use Whether the indi-ces for postural control developed in this research can beapplied to other tasks or postures must also be confirmedApplication software that facilitates therapistsrsquo quantitativeevaluations or verifications of the effectiveness of interven-tions would be especially important clinically Despite theselimitations this research advances the fusion of occupationalscience and practice by bringing together the fields of occu-pational therapy and CV In future works we are planningto focus on the clinical usability of the CV-based approachproposed in this research (eg user interface) Furthermorewe intend to develop a versatile quantification approach thatcan be used not only in standardised assessments but also inactual occupational activities and is applicable to variousdisorders The development of more precise quantificationmethods provides a basis for considering more effectiveinterventions and contributes to realizing occupational ther-apistsrsquo full potential

4 Conclusions

This study confirmed the close match between the results ofCV-based indices (SPB and AG) and TQCE Furthermorecompared with assessments using only the conventionalquantitative index (DT) evaluations that included CV-based indices provided more detailed quantitative informa-tion with lower time costs This research showed the possibil-ity of overcoming the problems associated with currentquantitative evaluations that they are insufficient for verify-ing clinical effects or detailing motor performances whichtherapists must observe qualitatively The results of this studyhave the following implications for occupational therapypractice

(i) Application of CV-based indices provides detailedquantitative and shareable information useful fornot only screening assessments but also for verifyinginterventions more precisely

(ii) CV-based approaches can reduce the time cost andthe physical and psychological burdens on clientsimposed by the performance of the assessment task

(iii) Leveraging CV technology enables the direct quanti-tative evaluation of motor performance duringactual occupational activities of daily living

This research also showed the possibility of making con-ventional evaluation tasks easier and simpler without specialequipment so that therapists can reduce time costs and thephysical and psychological burdens on clients This integra-

tion of occupational therapy and CV advances the scienceand practice of occupational therapy so that therapists canperform to their full potential

Data Availability

The data used to support the findings of this study are Excelfiles which consist of DT CV-based indices (SPB and AG)and TQCE of each participant Please contact the authorsof the article to address the data

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Authorsrsquo Contributions

H Hagihara and N Ienaga contributed equally to this work

Acknowledgments

We would like to thank T Kato and M Sakagami for theirinsightful comments and we are also grateful to our studyparticipants We used RAIDEN a supercomputer system ofAIP (RIKEN) to conduct pose estimation and analysis Thisresearch was supported by a Grant-in-Aid from the JapanSociety for the Promotion of Science (JSPS) Research Fellows(18J21948 and 17J05489)

References

[1] World Federation of Occupational Therapists Definitions ofoccupational therapy from member organisations WorldFederation of Occupational Therapists United Kingdom2019 July 2019 httpswwwwfotorgresourcesdefinitions-of-occupational-therapy-from-member-organisations

[2] A J Ayres Sensory Integration and the Child WesternPsychological Services Los Angeles (CA) 1979

[3] D Green T Charman A Pickles et al ldquoImpairment in move-ment skills of children with autistic spectrum disordersrdquoDevelopmental Medicine and Child Neurology vol 51 no 4pp 311ndash316 2009

[4] R Iwanaga C Kawasaki and R Tsuchida ldquoBrief report com-parison of sensory-motor and cognitive function betweenautism and Asperger syndrome in preschool childrenrdquo Journalof Autism and Developmental Disorders vol 30 no 2 pp 169ndash174 2000

[5] C P Whyatt and C M Craig ldquoMotor skills in children aged7ndash10 years diagnosed with autism spectrum disorderrdquo Journalof Autism and Developmental Disorders vol 42 no 9pp 1799ndash1809 2012

[6] J Konicarova P Bob and J Raboch ldquoBalance deficits andADHD symptoms in medication-naiumlve school-aged boysrdquoNeuropsychiatric Disease and Treatment vol 10 pp 85ndash882014

[7] M P Bucci C Stordeur E Acquaviva H Peyre andR Delorme ldquoPostural instability in children with ADHD isimproved by methylphenidaterdquo Frontiers in Neurosciencevol 10 p 163 2016

7Occupational Therapy International

key point for each frame is automatically calculated rangingfrom 0 to 1 (a higher value indicates higher confidence) sothat the degree of precision of key-point estimation can becomprehensible For example in Figure 1 all the key pointsused for subsequent analysis were detected with high confi-dence values (gt05)

OpenPose can detect the key points more accuratelywhen the head of a person is above and the legs below dueto the effect of the training data The input video was rotated90deg to turn the participantsrsquo heads upward since the partici-pants kept the bird dog posture as shown in Figure 1 Thekey points were then rotated 90deg back after being detectedby OpenPose To eliminate the detection noise key pointswhose confidence values were less than 05 were removedin each frame and interpolated linearly

Because occupational therapists evaluate SPB and AG inaddition to duration time we defined these items as shownbelow

254 Static Postural Balance (SPB) We first calculate themoving distance Dprime

D = 1n minus 1

nminus1

i=1

diiminus1bl

eth1THORN

Dprime = Dw +De +Da +Dketh THORN4 eth2THORN

D is the moving distance of a key point averaged in theframe direction (n represents the number of frames of thevideo) The moving distance diiminus1 between two consecutiveframes i i minus 1 is divided by the body length bl The bodylength is the average value of the distance between two keypoints corresponding to the neck and hip in all frames ofthe video The reason for dividing the moving distance bythe body length is to reduce the influences of the differencesin the heights of the subjects and in the distances between thecamera and the subjects There are actually four Ds for thewrist key point Dw and the elbow key point De of theextended arm and the ankle key point Da and the knee keypoint Dk of the extended leg SPB is the standardised value(z-score) obtained as follows

SPB =log 1Dprime

minus μ

σ eth3THORN

where μ and σ indicate the average and standard deviation oflog eth1DprimeTHORN respectively The average and standard deviationof SPB are 0 and 1 respectively because SPB is standardisedto eliminate the influence of numerical magnitude for multi-ple regression analysis with other indices High SPB indicateshigh static postural stability

255 Antigravity Posture (AG) AG was calculated as

a = minus1nnminus1

i=0φj j + θj jeth THORN eth4THORN

where jφj and jθj are the absolute angles between the floorand the extended arm and leg respectively as shown inFigure 1 Here ldquoarmrdquo refers to the vector connecting the neckposition to the average position of the wrist and elbow andldquolegrdquo refers to the vector connecting the hip position to theaverage position of the ankle and knee Because these areabsolute values the angle increases whether the limb is raisedor lowered a is the negative of the sum of the two anglesaveraged over the frame direction and AG is the standar-dised value (z-score) of a High AG indicates high antigravityposture

26 Data Analysis First we investigated how closely theCV-based scores (SPB and AG) reflected TQCE usingSpearmanrsquos rank correlation coefficient (analysis 1) Nextwe compared the effectiveness of the CV-based model tothat of the DT-based model (analysis 2) SPB AG andDT together or DT alone were used as the explanatoryvariables and regression analyses were performed usingTQCE as the objective variable with subsequent correla-tion analysis between the predicted value of the regressionanalysis and TQCE assuming that the test time wasunchanged

3 Results and Discussion

31 Degree to which CV-Based Scores Reflected TQCE Basedon the analyses using Spearmanrsquos rank correlation coeffi-cient SPB showed a significant positive correlation withTQCE (rs = 0654 p lt 0001) AG also showed a significantthough weak positive correlation with TQCE (rs = 0382p = 00013) TQCE tended to be higher when both of theseCV-based scores were high (Figure 2)

32 Multiple Regression Model Reflecting TQCE To comparethe evaluations using only the conventional quantitative index(DT) with the evaluations using all three indices (DT and theCV-based scores) we set TQCE as the objective variable andcalculated the model using DT as the explanatory variable ina single regression analysis in comparison with the modelusing SPB AG and DT as the explanatory variables in a mul-tiple regression analysis DT alone had a significant positive

ndash2

ndash2

ndash1

ndash1 0

0

1

AG

2

1 21

2

3

TQCE

4

5

SPB

Figure 2 Scatter plot of SPB and AG with colour bar of TQCE

4 Occupational Therapy International

correlation with TQCE (rs = 0886 p lt 0001) However thevalue calculated by multiple regression analysis whichincluded the CV-based scores showed a stronger significantpositive correlation with TQCE (rs = 0914 p lt 0001)

For test times shorter than 60 s the difference betweenthe CV- and DT-based scores was even more pronounced(Figure 3(a)) When the test time was reduced to 20 s theCV-based score showed a higher positive correlation

090

085

080

075

070

065

10 20 30 40Test time (sec)

CV-based scoreDT

50 60Sp

earm

anrsquos

rank

corr

elat

ion

coeffi

cien

t

(b)

(d)

(c)

(e)

(a)

CV-based score

rs = 0878

ndash2 ndash1 0

1

2

3TQCE

4

5

1

(b)

1

2

3TQCE

4

5

CV-based scorendash2 ndash1 0 1

rs = 0914

(c)

DT

1

5 10 15 20

2

3TQCE

4

5rs = 0789

(d)

DT0 20 40 60

1

2

3TQCE

4

5rs = 0886

(e)

Figure 3 Comparison of the CV- and DT-based models (a) Correlation coefficient between TQCE and CV- (blue) or DT-based (orange)models for every 10 s of test time (b) Scatter plot of CV-based scores and TQCE at the test time of 20 s (c) Scatter plot of CV-basedscores and TQCE at the test time of 60 s (d) Scatter plot of DT-based scores and TQCE at the test time of 20 s (e) Scatter plot of DT-based scores and TQCE at the test time of 60 s lowastlowastlowastp lt 0001

5Occupational Therapy International

(rs = 0878 p lt 0001 see Figures 3(b) and 3(c)) than DTalone (rs = 0789 p lt 0001 see Figures 3(d) and 3(e)) Inall 68 trials (34 participants times two sides) the number of trialsfor which DT was equal or more than 60 s was 18 (26) andthe number for which DT was equal or more than 20 swas 37 (54) The results of each regression analysis areshown in Table 1

33 Effectiveness of CV-Based Indices The results of thisresearch revealed the following two points First both SPBand AG had a significant positive correlation with TQCEand TQCE tended to be higher when both of these CV-based scores were high indicating that these indices quanti-fied TQCE well Second the multiple regression analysisand subsequent correlation analysis showed that the CV-based model reflected the TQCE to a greater degree thanthe conventional DT-based model indicating that evalua-tions using CV-based scores provide more detailed quantita-tive information These results showed the possibility ofovercoming the severe problems with current quantitativeevaluation indices which are useful for screening but arenot sufficient for verifying clinical effects or depicting detailsof motor performances including postural control Althougha certain proportion of children with developmental disor-ders have difficulties with postural control [3ndash9] they oftendemonstrate the maximum performance in terms of currentquantitative scores resulting in a ceiling effect [17] and theirdifficulties with postural control can only be identified bytherapistsrsquo clinical observations By using CV-based scorestherapists can quantitatively evaluate their postural controleven when they achieve maximumDT Although the correla-tion coefficient between AG and TQCE was significant butlow this result may have occurred because therapists evalu-ated AG posture only after confirming the achievement ofinitial evaluation items such as durability or static posturalbalance

The developed CV-based scores SPB and AG will con-tribute not only to identifying essential factors of posturalcontrol that have negative effects on childrenrsquos everydayactivities but also to establishing more valid evidence forthe efficacy of occupational therapy Body functions such aspostural control are the basis for childrenrsquos occupational per-formance on skills such as self-care communication andsubject learning [2] Furthermore children with develop-mental disorders especially developmental coordination dis-order often suffer from psychological problems including

lower self-esteem [35 36] anxiety and depression [37 38]The qualitative improvement of postural control might thuscontribute to improving their activities and participationand such improvements might be grasped by CV-basedscores because these detailed quantitative indices reflect thequalitative aspects of postural control These scores will alsomeet the need for developing functional outcome measureswhich are sensitive to occupational therapy interventions[10] Studies on the postural control of people with develop-mental disorders have focused on a specific posture onlysuch as standing on one leg (eg [39]) or on both legs (eg[40 41]) the results of which cannot be widely applied tovarious postures In contrast the CV-based approach willbroaden opportunities to advance detailed and quantitativeevaluations of the qualitative aspects of a broad range of pos-tures important for early postural development and activitiesin daily living

34 Applicability to the Development of Lower Cost and moreDetailed Assessments The degree to which the CV-basedmodel reflected TQCE was hardly reduced even when the testtime was shortened Furthermore the precision with whichthe models reflected TQCE were almost equal between theDT-based model at the test time of 60 s and the CV-basedmodel at the test time of 20 s This suggests the possibilityof making the current assessment task procedure easier andsimpler to reduce the time cost of testing and the physicaland psychological burden on clients In fact the high costand burden of conventional standardised assessments forchildren with poor motor skills are regarded as a clinicalproblem because longer test times cause fatigue and loss ofmotivation [17] The CV-based approach shortens the testtime and also yields more detailed quantitative informationthat is useful for both scientific research and clinical practice

The experimental task ldquoOne Arm and One Leg Balancerdquoused in this research simultaneously evaluates multipleaspects of postural control including at least SPB AG andDT Although developing several tasks to evaluate eachaspect separately may be desirable in particularly complexcases this increases the cost and physical and psychologicalburdens on patients and thus has poor clinical applicabilityThe research presented herein demonstrates that the pro-posed indices resolve and quantify multiple aspects of pos-tural control for multiple important aspects simultaneouslyoffering substantial reductions in the cost of the evaluationwithout compromising the amount of information

Table 1 Regression analysis results

Model using SPB AG and DT Model using only DTTest time 20 s Test time 60 s Test time 20 s Test time 60 s

Variables β (SE) β (SE) β (SE) β (SE)

SPB 035 (0083)lowastlowastlowast 025 (0059)lowastlowastlowast

AG 025 (0070)lowastlowastlowast 022 (0050)lowastlowastlowast

DT 050 (0068)lowastlowastlowast 069 (0060)lowastlowastlowast 076 (0081)lowastlowastlowast 088 (0059)lowastlowastlowast

Adjusted R2 069lowastlowastlowast 084lowastlowastlowast 057lowastlowastlowast 077lowastlowastlowast

lowastlowastlowastp lt 0001 therapistsrsquo qualitative clinical evaluation (TQCE) was set as the objective variable SPB static postural stability AG antigravity posture DTduration time SE standard error

6 Occupational Therapy International

35 Limitations and Future Work This research has severallimitations because we conducted a simple task on typicallydeveloping children so that the clinical applicability of CV-based quantification and analysis could be easily exploredA larger and randomized sample of children with and with-out developmental disorders must be beneficial to confirmthe validity of this CV-based approach Furthermore therelationship between the seriousness of motor skill impair-ment or its developmental change and CV-based indicesshould be investigated for diagnostic use Whether the indi-ces for postural control developed in this research can beapplied to other tasks or postures must also be confirmedApplication software that facilitates therapistsrsquo quantitativeevaluations or verifications of the effectiveness of interven-tions would be especially important clinically Despite theselimitations this research advances the fusion of occupationalscience and practice by bringing together the fields of occu-pational therapy and CV In future works we are planningto focus on the clinical usability of the CV-based approachproposed in this research (eg user interface) Furthermorewe intend to develop a versatile quantification approach thatcan be used not only in standardised assessments but also inactual occupational activities and is applicable to variousdisorders The development of more precise quantificationmethods provides a basis for considering more effectiveinterventions and contributes to realizing occupational ther-apistsrsquo full potential

4 Conclusions

This study confirmed the close match between the results ofCV-based indices (SPB and AG) and TQCE Furthermorecompared with assessments using only the conventionalquantitative index (DT) evaluations that included CV-based indices provided more detailed quantitative informa-tion with lower time costs This research showed the possibil-ity of overcoming the problems associated with currentquantitative evaluations that they are insufficient for verify-ing clinical effects or detailing motor performances whichtherapists must observe qualitatively The results of this studyhave the following implications for occupational therapypractice

(i) Application of CV-based indices provides detailedquantitative and shareable information useful fornot only screening assessments but also for verifyinginterventions more precisely

(ii) CV-based approaches can reduce the time cost andthe physical and psychological burdens on clientsimposed by the performance of the assessment task

(iii) Leveraging CV technology enables the direct quanti-tative evaluation of motor performance duringactual occupational activities of daily living

This research also showed the possibility of making con-ventional evaluation tasks easier and simpler without specialequipment so that therapists can reduce time costs and thephysical and psychological burdens on clients This integra-

tion of occupational therapy and CV advances the scienceand practice of occupational therapy so that therapists canperform to their full potential

Data Availability

The data used to support the findings of this study are Excelfiles which consist of DT CV-based indices (SPB and AG)and TQCE of each participant Please contact the authorsof the article to address the data

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Authorsrsquo Contributions

H Hagihara and N Ienaga contributed equally to this work

Acknowledgments

We would like to thank T Kato and M Sakagami for theirinsightful comments and we are also grateful to our studyparticipants We used RAIDEN a supercomputer system ofAIP (RIKEN) to conduct pose estimation and analysis Thisresearch was supported by a Grant-in-Aid from the JapanSociety for the Promotion of Science (JSPS) Research Fellows(18J21948 and 17J05489)

References

[1] World Federation of Occupational Therapists Definitions ofoccupational therapy from member organisations WorldFederation of Occupational Therapists United Kingdom2019 July 2019 httpswwwwfotorgresourcesdefinitions-of-occupational-therapy-from-member-organisations

[2] A J Ayres Sensory Integration and the Child WesternPsychological Services Los Angeles (CA) 1979

[3] D Green T Charman A Pickles et al ldquoImpairment in move-ment skills of children with autistic spectrum disordersrdquoDevelopmental Medicine and Child Neurology vol 51 no 4pp 311ndash316 2009

[4] R Iwanaga C Kawasaki and R Tsuchida ldquoBrief report com-parison of sensory-motor and cognitive function betweenautism and Asperger syndrome in preschool childrenrdquo Journalof Autism and Developmental Disorders vol 30 no 2 pp 169ndash174 2000

[5] C P Whyatt and C M Craig ldquoMotor skills in children aged7ndash10 years diagnosed with autism spectrum disorderrdquo Journalof Autism and Developmental Disorders vol 42 no 9pp 1799ndash1809 2012

[6] J Konicarova P Bob and J Raboch ldquoBalance deficits andADHD symptoms in medication-naiumlve school-aged boysrdquoNeuropsychiatric Disease and Treatment vol 10 pp 85ndash882014

[7] M P Bucci C Stordeur E Acquaviva H Peyre andR Delorme ldquoPostural instability in children with ADHD isimproved by methylphenidaterdquo Frontiers in Neurosciencevol 10 p 163 2016

7Occupational Therapy International

correlation with TQCE (rs = 0886 p lt 0001) However thevalue calculated by multiple regression analysis whichincluded the CV-based scores showed a stronger significantpositive correlation with TQCE (rs = 0914 p lt 0001)

For test times shorter than 60 s the difference betweenthe CV- and DT-based scores was even more pronounced(Figure 3(a)) When the test time was reduced to 20 s theCV-based score showed a higher positive correlation

090

085

080

075

070

065

10 20 30 40Test time (sec)

CV-based scoreDT

50 60Sp

earm

anrsquos

rank

corr

elat

ion

coeffi

cien

t

(b)

(d)

(c)

(e)

(a)

CV-based score

rs = 0878

ndash2 ndash1 0

1

2

3TQCE

4

5

1

(b)

1

2

3TQCE

4

5

CV-based scorendash2 ndash1 0 1

rs = 0914

(c)

DT

1

5 10 15 20

2

3TQCE

4

5rs = 0789

(d)

DT0 20 40 60

1

2

3TQCE

4

5rs = 0886

(e)

Figure 3 Comparison of the CV- and DT-based models (a) Correlation coefficient between TQCE and CV- (blue) or DT-based (orange)models for every 10 s of test time (b) Scatter plot of CV-based scores and TQCE at the test time of 20 s (c) Scatter plot of CV-basedscores and TQCE at the test time of 60 s (d) Scatter plot of DT-based scores and TQCE at the test time of 20 s (e) Scatter plot of DT-based scores and TQCE at the test time of 60 s lowastlowastlowastp lt 0001

5Occupational Therapy International

(rs = 0878 p lt 0001 see Figures 3(b) and 3(c)) than DTalone (rs = 0789 p lt 0001 see Figures 3(d) and 3(e)) Inall 68 trials (34 participants times two sides) the number of trialsfor which DT was equal or more than 60 s was 18 (26) andthe number for which DT was equal or more than 20 swas 37 (54) The results of each regression analysis areshown in Table 1

33 Effectiveness of CV-Based Indices The results of thisresearch revealed the following two points First both SPBand AG had a significant positive correlation with TQCEand TQCE tended to be higher when both of these CV-based scores were high indicating that these indices quanti-fied TQCE well Second the multiple regression analysisand subsequent correlation analysis showed that the CV-based model reflected the TQCE to a greater degree thanthe conventional DT-based model indicating that evalua-tions using CV-based scores provide more detailed quantita-tive information These results showed the possibility ofovercoming the severe problems with current quantitativeevaluation indices which are useful for screening but arenot sufficient for verifying clinical effects or depicting detailsof motor performances including postural control Althougha certain proportion of children with developmental disor-ders have difficulties with postural control [3ndash9] they oftendemonstrate the maximum performance in terms of currentquantitative scores resulting in a ceiling effect [17] and theirdifficulties with postural control can only be identified bytherapistsrsquo clinical observations By using CV-based scorestherapists can quantitatively evaluate their postural controleven when they achieve maximumDT Although the correla-tion coefficient between AG and TQCE was significant butlow this result may have occurred because therapists evalu-ated AG posture only after confirming the achievement ofinitial evaluation items such as durability or static posturalbalance

The developed CV-based scores SPB and AG will con-tribute not only to identifying essential factors of posturalcontrol that have negative effects on childrenrsquos everydayactivities but also to establishing more valid evidence forthe efficacy of occupational therapy Body functions such aspostural control are the basis for childrenrsquos occupational per-formance on skills such as self-care communication andsubject learning [2] Furthermore children with develop-mental disorders especially developmental coordination dis-order often suffer from psychological problems including

lower self-esteem [35 36] anxiety and depression [37 38]The qualitative improvement of postural control might thuscontribute to improving their activities and participationand such improvements might be grasped by CV-basedscores because these detailed quantitative indices reflect thequalitative aspects of postural control These scores will alsomeet the need for developing functional outcome measureswhich are sensitive to occupational therapy interventions[10] Studies on the postural control of people with develop-mental disorders have focused on a specific posture onlysuch as standing on one leg (eg [39]) or on both legs (eg[40 41]) the results of which cannot be widely applied tovarious postures In contrast the CV-based approach willbroaden opportunities to advance detailed and quantitativeevaluations of the qualitative aspects of a broad range of pos-tures important for early postural development and activitiesin daily living

34 Applicability to the Development of Lower Cost and moreDetailed Assessments The degree to which the CV-basedmodel reflected TQCE was hardly reduced even when the testtime was shortened Furthermore the precision with whichthe models reflected TQCE were almost equal between theDT-based model at the test time of 60 s and the CV-basedmodel at the test time of 20 s This suggests the possibilityof making the current assessment task procedure easier andsimpler to reduce the time cost of testing and the physicaland psychological burden on clients In fact the high costand burden of conventional standardised assessments forchildren with poor motor skills are regarded as a clinicalproblem because longer test times cause fatigue and loss ofmotivation [17] The CV-based approach shortens the testtime and also yields more detailed quantitative informationthat is useful for both scientific research and clinical practice

The experimental task ldquoOne Arm and One Leg Balancerdquoused in this research simultaneously evaluates multipleaspects of postural control including at least SPB AG andDT Although developing several tasks to evaluate eachaspect separately may be desirable in particularly complexcases this increases the cost and physical and psychologicalburdens on patients and thus has poor clinical applicabilityThe research presented herein demonstrates that the pro-posed indices resolve and quantify multiple aspects of pos-tural control for multiple important aspects simultaneouslyoffering substantial reductions in the cost of the evaluationwithout compromising the amount of information

Table 1 Regression analysis results

Model using SPB AG and DT Model using only DTTest time 20 s Test time 60 s Test time 20 s Test time 60 s

Variables β (SE) β (SE) β (SE) β (SE)

SPB 035 (0083)lowastlowastlowast 025 (0059)lowastlowastlowast

AG 025 (0070)lowastlowastlowast 022 (0050)lowastlowastlowast

DT 050 (0068)lowastlowastlowast 069 (0060)lowastlowastlowast 076 (0081)lowastlowastlowast 088 (0059)lowastlowastlowast

Adjusted R2 069lowastlowastlowast 084lowastlowastlowast 057lowastlowastlowast 077lowastlowastlowast

lowastlowastlowastp lt 0001 therapistsrsquo qualitative clinical evaluation (TQCE) was set as the objective variable SPB static postural stability AG antigravity posture DTduration time SE standard error

6 Occupational Therapy International

35 Limitations and Future Work This research has severallimitations because we conducted a simple task on typicallydeveloping children so that the clinical applicability of CV-based quantification and analysis could be easily exploredA larger and randomized sample of children with and with-out developmental disorders must be beneficial to confirmthe validity of this CV-based approach Furthermore therelationship between the seriousness of motor skill impair-ment or its developmental change and CV-based indicesshould be investigated for diagnostic use Whether the indi-ces for postural control developed in this research can beapplied to other tasks or postures must also be confirmedApplication software that facilitates therapistsrsquo quantitativeevaluations or verifications of the effectiveness of interven-tions would be especially important clinically Despite theselimitations this research advances the fusion of occupationalscience and practice by bringing together the fields of occu-pational therapy and CV In future works we are planningto focus on the clinical usability of the CV-based approachproposed in this research (eg user interface) Furthermorewe intend to develop a versatile quantification approach thatcan be used not only in standardised assessments but also inactual occupational activities and is applicable to variousdisorders The development of more precise quantificationmethods provides a basis for considering more effectiveinterventions and contributes to realizing occupational ther-apistsrsquo full potential

4 Conclusions

This study confirmed the close match between the results ofCV-based indices (SPB and AG) and TQCE Furthermorecompared with assessments using only the conventionalquantitative index (DT) evaluations that included CV-based indices provided more detailed quantitative informa-tion with lower time costs This research showed the possibil-ity of overcoming the problems associated with currentquantitative evaluations that they are insufficient for verify-ing clinical effects or detailing motor performances whichtherapists must observe qualitatively The results of this studyhave the following implications for occupational therapypractice

(i) Application of CV-based indices provides detailedquantitative and shareable information useful fornot only screening assessments but also for verifyinginterventions more precisely

(ii) CV-based approaches can reduce the time cost andthe physical and psychological burdens on clientsimposed by the performance of the assessment task

(iii) Leveraging CV technology enables the direct quanti-tative evaluation of motor performance duringactual occupational activities of daily living

This research also showed the possibility of making con-ventional evaluation tasks easier and simpler without specialequipment so that therapists can reduce time costs and thephysical and psychological burdens on clients This integra-

tion of occupational therapy and CV advances the scienceand practice of occupational therapy so that therapists canperform to their full potential

Data Availability

The data used to support the findings of this study are Excelfiles which consist of DT CV-based indices (SPB and AG)and TQCE of each participant Please contact the authorsof the article to address the data

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Authorsrsquo Contributions

H Hagihara and N Ienaga contributed equally to this work

Acknowledgments

We would like to thank T Kato and M Sakagami for theirinsightful comments and we are also grateful to our studyparticipants We used RAIDEN a supercomputer system ofAIP (RIKEN) to conduct pose estimation and analysis Thisresearch was supported by a Grant-in-Aid from the JapanSociety for the Promotion of Science (JSPS) Research Fellows(18J21948 and 17J05489)

References

[1] World Federation of Occupational Therapists Definitions ofoccupational therapy from member organisations WorldFederation of Occupational Therapists United Kingdom2019 July 2019 httpswwwwfotorgresourcesdefinitions-of-occupational-therapy-from-member-organisations

[2] A J Ayres Sensory Integration and the Child WesternPsychological Services Los Angeles (CA) 1979

[3] D Green T Charman A Pickles et al ldquoImpairment in move-ment skills of children with autistic spectrum disordersrdquoDevelopmental Medicine and Child Neurology vol 51 no 4pp 311ndash316 2009

[4] R Iwanaga C Kawasaki and R Tsuchida ldquoBrief report com-parison of sensory-motor and cognitive function betweenautism and Asperger syndrome in preschool childrenrdquo Journalof Autism and Developmental Disorders vol 30 no 2 pp 169ndash174 2000

[5] C P Whyatt and C M Craig ldquoMotor skills in children aged7ndash10 years diagnosed with autism spectrum disorderrdquo Journalof Autism and Developmental Disorders vol 42 no 9pp 1799ndash1809 2012

[6] J Konicarova P Bob and J Raboch ldquoBalance deficits andADHD symptoms in medication-naiumlve school-aged boysrdquoNeuropsychiatric Disease and Treatment vol 10 pp 85ndash882014

[7] M P Bucci C Stordeur E Acquaviva H Peyre andR Delorme ldquoPostural instability in children with ADHD isimproved by methylphenidaterdquo Frontiers in Neurosciencevol 10 p 163 2016

7Occupational Therapy International

(rs = 0878 p lt 0001 see Figures 3(b) and 3(c)) than DTalone (rs = 0789 p lt 0001 see Figures 3(d) and 3(e)) Inall 68 trials (34 participants times two sides) the number of trialsfor which DT was equal or more than 60 s was 18 (26) andthe number for which DT was equal or more than 20 swas 37 (54) The results of each regression analysis areshown in Table 1

33 Effectiveness of CV-Based Indices The results of thisresearch revealed the following two points First both SPBand AG had a significant positive correlation with TQCEand TQCE tended to be higher when both of these CV-based scores were high indicating that these indices quanti-fied TQCE well Second the multiple regression analysisand subsequent correlation analysis showed that the CV-based model reflected the TQCE to a greater degree thanthe conventional DT-based model indicating that evalua-tions using CV-based scores provide more detailed quantita-tive information These results showed the possibility ofovercoming the severe problems with current quantitativeevaluation indices which are useful for screening but arenot sufficient for verifying clinical effects or depicting detailsof motor performances including postural control Althougha certain proportion of children with developmental disor-ders have difficulties with postural control [3ndash9] they oftendemonstrate the maximum performance in terms of currentquantitative scores resulting in a ceiling effect [17] and theirdifficulties with postural control can only be identified bytherapistsrsquo clinical observations By using CV-based scorestherapists can quantitatively evaluate their postural controleven when they achieve maximumDT Although the correla-tion coefficient between AG and TQCE was significant butlow this result may have occurred because therapists evalu-ated AG posture only after confirming the achievement ofinitial evaluation items such as durability or static posturalbalance

The developed CV-based scores SPB and AG will con-tribute not only to identifying essential factors of posturalcontrol that have negative effects on childrenrsquos everydayactivities but also to establishing more valid evidence forthe efficacy of occupational therapy Body functions such aspostural control are the basis for childrenrsquos occupational per-formance on skills such as self-care communication andsubject learning [2] Furthermore children with develop-mental disorders especially developmental coordination dis-order often suffer from psychological problems including

lower self-esteem [35 36] anxiety and depression [37 38]The qualitative improvement of postural control might thuscontribute to improving their activities and participationand such improvements might be grasped by CV-basedscores because these detailed quantitative indices reflect thequalitative aspects of postural control These scores will alsomeet the need for developing functional outcome measureswhich are sensitive to occupational therapy interventions[10] Studies on the postural control of people with develop-mental disorders have focused on a specific posture onlysuch as standing on one leg (eg [39]) or on both legs (eg[40 41]) the results of which cannot be widely applied tovarious postures In contrast the CV-based approach willbroaden opportunities to advance detailed and quantitativeevaluations of the qualitative aspects of a broad range of pos-tures important for early postural development and activitiesin daily living

34 Applicability to the Development of Lower Cost and moreDetailed Assessments The degree to which the CV-basedmodel reflected TQCE was hardly reduced even when the testtime was shortened Furthermore the precision with whichthe models reflected TQCE were almost equal between theDT-based model at the test time of 60 s and the CV-basedmodel at the test time of 20 s This suggests the possibilityof making the current assessment task procedure easier andsimpler to reduce the time cost of testing and the physicaland psychological burden on clients In fact the high costand burden of conventional standardised assessments forchildren with poor motor skills are regarded as a clinicalproblem because longer test times cause fatigue and loss ofmotivation [17] The CV-based approach shortens the testtime and also yields more detailed quantitative informationthat is useful for both scientific research and clinical practice

The experimental task ldquoOne Arm and One Leg Balancerdquoused in this research simultaneously evaluates multipleaspects of postural control including at least SPB AG andDT Although developing several tasks to evaluate eachaspect separately may be desirable in particularly complexcases this increases the cost and physical and psychologicalburdens on patients and thus has poor clinical applicabilityThe research presented herein demonstrates that the pro-posed indices resolve and quantify multiple aspects of pos-tural control for multiple important aspects simultaneouslyoffering substantial reductions in the cost of the evaluationwithout compromising the amount of information

Table 1 Regression analysis results

Model using SPB AG and DT Model using only DTTest time 20 s Test time 60 s Test time 20 s Test time 60 s

Variables β (SE) β (SE) β (SE) β (SE)

SPB 035 (0083)lowastlowastlowast 025 (0059)lowastlowastlowast

AG 025 (0070)lowastlowastlowast 022 (0050)lowastlowastlowast

DT 050 (0068)lowastlowastlowast 069 (0060)lowastlowastlowast 076 (0081)lowastlowastlowast 088 (0059)lowastlowastlowast

Adjusted R2 069lowastlowastlowast 084lowastlowastlowast 057lowastlowastlowast 077lowastlowastlowast

lowastlowastlowastp lt 0001 therapistsrsquo qualitative clinical evaluation (TQCE) was set as the objective variable SPB static postural stability AG antigravity posture DTduration time SE standard error

6 Occupational Therapy International

35 Limitations and Future Work This research has severallimitations because we conducted a simple task on typicallydeveloping children so that the clinical applicability of CV-based quantification and analysis could be easily exploredA larger and randomized sample of children with and with-out developmental disorders must be beneficial to confirmthe validity of this CV-based approach Furthermore therelationship between the seriousness of motor skill impair-ment or its developmental change and CV-based indicesshould be investigated for diagnostic use Whether the indi-ces for postural control developed in this research can beapplied to other tasks or postures must also be confirmedApplication software that facilitates therapistsrsquo quantitativeevaluations or verifications of the effectiveness of interven-tions would be especially important clinically Despite theselimitations this research advances the fusion of occupationalscience and practice by bringing together the fields of occu-pational therapy and CV In future works we are planningto focus on the clinical usability of the CV-based approachproposed in this research (eg user interface) Furthermorewe intend to develop a versatile quantification approach thatcan be used not only in standardised assessments but also inactual occupational activities and is applicable to variousdisorders The development of more precise quantificationmethods provides a basis for considering more effectiveinterventions and contributes to realizing occupational ther-apistsrsquo full potential

4 Conclusions

This study confirmed the close match between the results ofCV-based indices (SPB and AG) and TQCE Furthermorecompared with assessments using only the conventionalquantitative index (DT) evaluations that included CV-based indices provided more detailed quantitative informa-tion with lower time costs This research showed the possibil-ity of overcoming the problems associated with currentquantitative evaluations that they are insufficient for verify-ing clinical effects or detailing motor performances whichtherapists must observe qualitatively The results of this studyhave the following implications for occupational therapypractice

(i) Application of CV-based indices provides detailedquantitative and shareable information useful fornot only screening assessments but also for verifyinginterventions more precisely

(ii) CV-based approaches can reduce the time cost andthe physical and psychological burdens on clientsimposed by the performance of the assessment task

(iii) Leveraging CV technology enables the direct quanti-tative evaluation of motor performance duringactual occupational activities of daily living

This research also showed the possibility of making con-ventional evaluation tasks easier and simpler without specialequipment so that therapists can reduce time costs and thephysical and psychological burdens on clients This integra-

tion of occupational therapy and CV advances the scienceand practice of occupational therapy so that therapists canperform to their full potential

Data Availability

The data used to support the findings of this study are Excelfiles which consist of DT CV-based indices (SPB and AG)and TQCE of each participant Please contact the authorsof the article to address the data

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Authorsrsquo Contributions

H Hagihara and N Ienaga contributed equally to this work

Acknowledgments

We would like to thank T Kato and M Sakagami for theirinsightful comments and we are also grateful to our studyparticipants We used RAIDEN a supercomputer system ofAIP (RIKEN) to conduct pose estimation and analysis Thisresearch was supported by a Grant-in-Aid from the JapanSociety for the Promotion of Science (JSPS) Research Fellows(18J21948 and 17J05489)

References

[1] World Federation of Occupational Therapists Definitions ofoccupational therapy from member organisations WorldFederation of Occupational Therapists United Kingdom2019 July 2019 httpswwwwfotorgresourcesdefinitions-of-occupational-therapy-from-member-organisations

[2] A J Ayres Sensory Integration and the Child WesternPsychological Services Los Angeles (CA) 1979

[3] D Green T Charman A Pickles et al ldquoImpairment in move-ment skills of children with autistic spectrum disordersrdquoDevelopmental Medicine and Child Neurology vol 51 no 4pp 311ndash316 2009

[4] R Iwanaga C Kawasaki and R Tsuchida ldquoBrief report com-parison of sensory-motor and cognitive function betweenautism and Asperger syndrome in preschool childrenrdquo Journalof Autism and Developmental Disorders vol 30 no 2 pp 169ndash174 2000

[5] C P Whyatt and C M Craig ldquoMotor skills in children aged7ndash10 years diagnosed with autism spectrum disorderrdquo Journalof Autism and Developmental Disorders vol 42 no 9pp 1799ndash1809 2012

[6] J Konicarova P Bob and J Raboch ldquoBalance deficits andADHD symptoms in medication-naiumlve school-aged boysrdquoNeuropsychiatric Disease and Treatment vol 10 pp 85ndash882014

[7] M P Bucci C Stordeur E Acquaviva H Peyre andR Delorme ldquoPostural instability in children with ADHD isimproved by methylphenidaterdquo Frontiers in Neurosciencevol 10 p 163 2016

7Occupational Therapy International

35 Limitations and Future Work This research has severallimitations because we conducted a simple task on typicallydeveloping children so that the clinical applicability of CV-based quantification and analysis could be easily exploredA larger and randomized sample of children with and with-out developmental disorders must be beneficial to confirmthe validity of this CV-based approach Furthermore therelationship between the seriousness of motor skill impair-ment or its developmental change and CV-based indicesshould be investigated for diagnostic use Whether the indi-ces for postural control developed in this research can beapplied to other tasks or postures must also be confirmedApplication software that facilitates therapistsrsquo quantitativeevaluations or verifications of the effectiveness of interven-tions would be especially important clinically Despite theselimitations this research advances the fusion of occupationalscience and practice by bringing together the fields of occu-pational therapy and CV In future works we are planningto focus on the clinical usability of the CV-based approachproposed in this research (eg user interface) Furthermorewe intend to develop a versatile quantification approach thatcan be used not only in standardised assessments but also inactual occupational activities and is applicable to variousdisorders The development of more precise quantificationmethods provides a basis for considering more effectiveinterventions and contributes to realizing occupational ther-apistsrsquo full potential

4 Conclusions

This study confirmed the close match between the results ofCV-based indices (SPB and AG) and TQCE Furthermorecompared with assessments using only the conventionalquantitative index (DT) evaluations that included CV-based indices provided more detailed quantitative informa-tion with lower time costs This research showed the possibil-ity of overcoming the problems associated with currentquantitative evaluations that they are insufficient for verify-ing clinical effects or detailing motor performances whichtherapists must observe qualitatively The results of this studyhave the following implications for occupational therapypractice

(i) Application of CV-based indices provides detailedquantitative and shareable information useful fornot only screening assessments but also for verifyinginterventions more precisely

(ii) CV-based approaches can reduce the time cost andthe physical and psychological burdens on clientsimposed by the performance of the assessment task

(iii) Leveraging CV technology enables the direct quanti-tative evaluation of motor performance duringactual occupational activities of daily living

This research also showed the possibility of making con-ventional evaluation tasks easier and simpler without specialequipment so that therapists can reduce time costs and thephysical and psychological burdens on clients This integra-

tion of occupational therapy and CV advances the scienceand practice of occupational therapy so that therapists canperform to their full potential

Data Availability

The data used to support the findings of this study are Excelfiles which consist of DT CV-based indices (SPB and AG)and TQCE of each participant Please contact the authorsof the article to address the data

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Authorsrsquo Contributions

H Hagihara and N Ienaga contributed equally to this work

Acknowledgments

We would like to thank T Kato and M Sakagami for theirinsightful comments and we are also grateful to our studyparticipants We used RAIDEN a supercomputer system ofAIP (RIKEN) to conduct pose estimation and analysis Thisresearch was supported by a Grant-in-Aid from the JapanSociety for the Promotion of Science (JSPS) Research Fellows(18J21948 and 17J05489)

References

[1] World Federation of Occupational Therapists Definitions ofoccupational therapy from member organisations WorldFederation of Occupational Therapists United Kingdom2019 July 2019 httpswwwwfotorgresourcesdefinitions-of-occupational-therapy-from-member-organisations

[2] A J Ayres Sensory Integration and the Child WesternPsychological Services Los Angeles (CA) 1979

[3] D Green T Charman A Pickles et al ldquoImpairment in move-ment skills of children with autistic spectrum disordersrdquoDevelopmental Medicine and Child Neurology vol 51 no 4pp 311ndash316 2009

[4] R Iwanaga C Kawasaki and R Tsuchida ldquoBrief report com-parison of sensory-motor and cognitive function betweenautism and Asperger syndrome in preschool childrenrdquo Journalof Autism and Developmental Disorders vol 30 no 2 pp 169ndash174 2000

[5] C P Whyatt and C M Craig ldquoMotor skills in children aged7ndash10 years diagnosed with autism spectrum disorderrdquo Journalof Autism and Developmental Disorders vol 42 no 9pp 1799ndash1809 2012

[6] J Konicarova P Bob and J Raboch ldquoBalance deficits andADHD symptoms in medication-naiumlve school-aged boysrdquoNeuropsychiatric Disease and Treatment vol 10 pp 85ndash882014

[7] M P Bucci C Stordeur E Acquaviva H Peyre andR Delorme ldquoPostural instability in children with ADHD isimproved by methylphenidaterdquo Frontiers in Neurosciencevol 10 p 163 2016

7Occupational Therapy International

[8] R J Cherng Y W Hsu Y J Chen and J Y Chen ldquoStandingbalance of children with developmental coordination disorderunder altered sensory conditionsrdquo Human Movement Sciencevol 26 no 6 pp 913ndash926 2007

[9] S S M Fong V Y L Lee and M Y C Pang ldquoSensory orga-nization of balance control in children with developmentalcoordination disorderrdquo Research in Developmental Disabil-ities vol 32 no 6 pp 2376ndash2382 2011

[10] J C Rogers and M B Holm ldquoAccepting the challenge of out-come research examining the effectiveness of occupationaltherapy practicerdquo The American Journal of Occupational Ther-apy vol 48 no 10 pp 871ndash876 1994

[11] S E Henderson D Sugden and A L Barnett MovementAssessment Battery for Children-2 The Psychological Corpora-tion London 2007

[12] R H Bruininks and B D Bruininks Bruininks-Oseretsky Testof Motor Proficiency Second Edition (BOT-2) Pearson Assess-ment Minneapolis MN 2005

[13] Japanese Academy of Sensory Integration The Japanese Play-ful Assessment for Neuropsychological Abilities ExaminerrsquosManual Pacific Supply Osaka 2011

[14] C DeMatteo M Law D Russell N Pollock P Rosenbaumand S Walter ldquoThe reliability and validity of the quality ofupper extremity skills testrdquo Physical amp Occupational TherapyIn Pediatrics vol 13 no 2 pp 1ndash18 1993

[15] B Bonnechegravere B Jansen and S van Sint Jan ldquoCost-effective(gaming) motion and balance devices for functional assess-ment need or hyperdquo Journal of Biomechanics vol 49no 13 pp 2561ndash2565 2016

[16] J L Alberts J R Hirsch M M Koop et al ldquoUsingaccelerometer and gyroscopic measures to quantify posturalstabilityrdquo Journal of Athletic Training vol 50 no 6pp 578ndash588 2015

[17] B French N J Sycamore H L McGlashan C C VBlanchard and N P Holmes ldquoCeiling effects in the move-ment assessment battery for children-2 (MABC-2) suggestthat non-parametric scoring methods are requiredrdquo PLoSOne vol 13 no 6 2018

[18] J S Sellers ldquoRelationship between antigravity control and pos-tural control in young childrenrdquo Physical Therapy vol 68no 4 pp 486ndash490 1988

[19] N A Odunaiya O M Oladeji and O O Oguntibeju ldquoAssess-ment of antigravity and postural control in health children inIbadan Nigeriardquo Pakistan Journal of Medica Sciencesvol 25 pp 583ndash590 2009

[20] Z Cao T Simon S E Wei and Y Sheikh ldquoRealtime multi-person 2D pose estimation using part affinity fieldsrdquo in Pro-ceedings of the IEEE Conference on Computer Vision and Pat-tern Recognition USA 2017Honolulu (HI) IEEE

[21] I Habibie W Xu D Mehta G Pons-Moll and C TheobaltldquoIn the wild human pose estimation using explicit 2D featuresand intermediate 3D representationsrdquo in Proceedings of theIEEE Conference on Computer Vision and Pattern Recognitionpp 10905ndash10914 2019

[22] W Yang W Ouyang X Wang J Ren H Li and X Wangldquo3D human pose estimation in the wild by adversarial learn-ingrdquo in Proceedings of the IEEE Conference on ComputerVision and Pattern Recognition pp 5255ndash5264 2018

[23] K He G Gkioxari P Dollaacuter and R Girshick ldquoMask r-cnnrdquoin Proceedings of the IEEE international conference on com-puter vision pp 2961ndash2969 2017

[24] C Chen Y Zhuang and J Xiao ldquoSilhouette representationand matching for 3D pose discrimination - A comparativestudyrdquo Image and Vision Computing vol 28 no 4 pp 654ndash667 2010

[25] D Xue A Sayana E Darke et al ldquoVision-based gait analysisfor senior carerdquo 2018 httparxivorgabs181200169

[26] H Chen X Luo Z Zheng and J Ke ldquoA proactive workerssafety risk evaluation framework based on position and pos-ture data fusionrdquo Automation in Construction vol 98pp 275ndash288 2019

[27] K Campbell K L H Carpenter J Hashemi et al ldquoComputervision analysis captures atypical attention in toddlers withautismrdquo Autism vol 23 no 3 pp 619ndash628 2018

[28] G Dawson K Campbell J Hashemi et al ldquoAtypical posturalcontrol can be detected via computer vision analysis in tod-dlers with autism spectrum disorderrdquo Scientific Reportsvol 8 no 1 2018

[29] L Tao A Paiement D Damen et al ldquoA comparative study ofpose representation and dynamics modelling for onlinemotion quality assessmentrdquo Computer Vision and ImageUnderstanding vol 148 pp 136ndash152 2016

[30] R Iwanaga A Ohta T Kato R Tsuchida K Hida andT Yamada ldquoDevelopment of Japanese playful assessment ofneuropsychological abilities correlations between test scoresand age in equilibrium tests antigravity posture tests andsomatosensory testsrdquo in Proceedings of the 15th World Con-gress of the World Federation of Occupational Therapists San-tiago Chile 2010WFOT

[31] T Kato K Hida R Iwanaga et al ldquoDevelopment of Japaneseplayful assessment of neuropsychological abilities correlationsbetween test scores and age in eye-hand coordination andvisual perception tests and praxis testsrdquo in Proceedings of the15th World Congress of the World Federation of OccupationalTherapists Santiago Chile 2010WFOT

[32] A J Ayres Sensory Integration and Praxis Test SIPT ManualWestern Psychological Services Los Angeles (CA) 1989

[33] T Kato R Iwanaga A Ota et al ldquoReliability of the Japaneseplayful assessment for neuropsychological abilities (JPAN)rdquoJapanese Journal of Sensory Integration vol 15 pp 19ndash242015

[34] Y Miyake R Kobayashi D Kelepecz andM Nakajima ldquoCoreexercises elevate trunk stability to facilitate skilled motorbehavior of the upper extremitiesrdquo Journal of Bodywork andMovement Therapies vol 17 no 2 pp 259ndash265 2013

[35] S McWilliams ldquoDevelopmental coordination disorder andself-esteem do occupational therapy groups have a positiveeffectrdquo British Journal of Occupational Therapy vol 68no 9 pp 393ndash400 2005

[36] A A Poulsen and J M Ziviani ldquoCan I play too Physical activ-ity engagement of children with developmental coordinationdisordersrdquo Canadian Journal of Occupational Therapyvol 71 no 2 pp 100ndash107 2004

[37] C Missiuna J Cairney N Pollock et al ldquoPsychological dis-tress in children with developmental coordination disorderand attention-deficit hyperactivity disorderrdquo Research inDevelopmental Disabilities vol 35 no 5 pp 1198ndash1207 2014

[38] J P Piek D Rigoli J G Pearsall-Jones et al ldquoDepressivesymptomatology in child and adolescent twins withattention-deficit hyperactivity disorder andor developmentalcoordination disorderrdquo Twin Research and Human Geneticsvol 10 no 4 pp 587ndash596 2007

8 Occupational Therapy International

[39] C L Tsai S K Wu and C H Huang ldquoStatic balance in chil-dren with developmental coordination disorderrdquo HumanMovement Science vol 27 no 1 pp 142ndash153 2008

[40] M Doumas R McKenna and B Murphy ldquoPostural controldeficits in autism spectrum disorder the role of sensory inte-grationrdquo Journal of Autism and Developmental Disordersvol 46 no 3 pp 853ndash861 2016

[41] Y H Lim K Partridge S Girdler and S L Morris ldquoStandingpostural control in individuals with autism spectrum disordersystematic review and meta-analysisrdquo Journal of Autism andDevelopmental Disorders vol 47 no 7 pp 2238ndash2253 2017

9Occupational Therapy International

• Computer Vision-Based Approach for Quantifying Occupational Therapistsrsquo Qualitative Evaluations of Postural Control
• 1 Introduction
• 2 Materials and Methods
• • 21 Design
  • 22 Participants
  • 23 Instrumentation
  • 24 Task Procedure
  • 25 Applied Indices
  • • 251 DT
    • 252 TQCE
    • 253 CV-Based Approach
    • 254 Static Postural Balance (SPB)
    • 255 Antigravity Posture (AG)
    • • 26 Data Analysis
      • • 3 Results and Discussion
        • • 31 Degree to which CV-Based Scores Reflected TQCE
          • 32 Multiple Regression Model Reflecting TQCE
          • 33 Effectiveness of CV-Based Indices
          • 34 Applicability to the Development of Lower Cost and more Detailed Assessments
          • 35 Limitations and Future Work
          • • 4 Conclusions
            • Data Availability
            • Conflicts of Interest
            • Authorsrsquo Contributions
            • Acknowledgments