A Random Forest Approach for Quantifying Gait Ataxia With Truncal and Peripheral Measurements Using Multiple Wearable Sensors

Phan, Dung H.; Nguyen, Nhan; Pathirana, Pubudu N.; Horne, Malcolm; Power, Laura; Szmulewicz, David J.

doi:10.1109/jsen.2019.2943879

Cited by 29 publications

(32 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our overall aim is to construct superior tests for cerebellar disease for use by non-subspecialist clinicians; for deep phenotyping and identifying of biomarkers for possible CA treatment trials. In earlier studies [4][5][6][7][8][9][10][11] , we demonstrated that our instrumented devices were able to accurately emulate each of the individual bedside tests of cerebellar function and to go beyond this by grading impairment severity. The research outlined in this paper aimed to produce a composite score of CA (based on a reduced number of our instrumented tests).…”

mentioning

confidence: 73%

“…The objective assessments of the nine neurological tests are grouped into the following 5 domains with a brief description of their execution protocol. A summary of features that proved pivotal in the diagnosis of CA in the related respective study 4,[6][7][8][9][10][11] has also been included for each test subsection in Table 2.…”

Section: Data Analysis Results Are Transformed Into a Clinically Relementioning

confidence: 99%

“…While wearable sensors are increasingly being explored as measurement tools for disability in neurological diseases like Parkinson's disease 12,13 and MS 14 , studies are still lacking in CA 15,16 with the potential of delivering a comprehensive, quantifiable, objective and meaningful measure of neurological function covering the 5 domains-speech, upper limb, lower limb, balance and gait. Our earlier studies [4][5][6][7][8][9][10][11] examined data from wearable sensors while subjects with CA and normal subjects (controls) performed one or more of the standard bedside tests for CA, pertaining to any one of the 5 domains. Our aim was to use the features extracted from this data to build statistical models that could accurately predict the presence and severity of CA as measured by the SARA and other similar scales.…”

mentioning

confidence: 99%

“…In the present study we used the results obtained in the previous CA assessments in speech 4,6 , upper limb [7][8][9] , lower limb 7,8 , balance 10 and gait 11 domains. The aim was to investigate differences between controls and CA subjects.…”

mentioning

confidence: 99%

See 3 more Smart Citations

Objective Assessment of Cerebellar Ataxia: A Comprehensive and Refined Approach

Kashyap

Phan

Pathirana

et al. 2020

Sci Rep

Self Cite

View full text Add to dashboard Cite

parametric analysis of cerebellar Ataxia (cA) could be of immense value compared to its subjective clinical assessments. this study focuses on a comprehensive scheme for objective assessment of cA through the instrumented versions of 9 commonly used neurological tests in 5 domains-speech, upper limb, lower limb, gait and balance. Twenty-three individuals diagnosed with CA to varying degrees and eleven age-matched healthy controls were recruited. Wearable inertial sensors and Kinect camera were utilised for data acquisition. Binary and multilabel discrimination power and intra-domain relationships of the features extracted from the sensor measures and the clinical scores were compared using Graph Theory, Centrality Measures, Random Forest binary and multilabel classification approaches. An optimal subset of 13 most important Principal Component (PC) features were selected for CAcontrol classification. This classification model resulted in an impressive performance accuracy of 97% (F1 score = 95.2%) with Holmesian dimensions distributed as 47.7% Stability, 6.3% Timing, 38.75% Accuracy and 7.24% Rhythmicity. Another optimal subset of 11 PC features demonstrated an F1 score of 84.2% in mapping the total 27 PC across 5 domains during CA multilabel discrimination. In both cases, the balance (Romberg) test contributed the most (31.1% and 42% respectively), followed by the peripheral tests whereas gait (Walking) test contributed the least. These findings paved the way for a better understanding of the feasibility of an instrumented system to assist informed clinical decisionmaking.

show abstract

mentioning

confidence: 73%

Section: Data Analysis Results Are Transformed Into a Clinically Relementioning

confidence: 99%

mentioning

confidence: 99%

mentioning

confidence: 99%

See 2 more Smart Citations

Objective Assessment of Cerebellar Ataxia: A Comprehensive and Refined Approach

Kashyap

Phan

Pathirana

et al. 2020

Sci Rep

Self Cite

View full text Add to dashboard Cite

show abstract

“…To overcome this and improve the prediction power the number of features were reduced using feature selection (FS) techniques. The Feed Backward Feature Elimination (FBE) [25] was employed along with three other widely-used methods involving Random Forest [26], RELIEF [27] and LASSO [28]. The central idea of FBE is to find a subset of features that increases the model's performance.…”

Section: Feature Selectionmentioning

confidence: 99%

A Comprehensive Scheme for the Objective Upper Body Assessments of Subjects with Cerebellar Ataxia

Tran

Nguyen

Pathirana

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

BackgroundCerebellar Ataxia refers to the disturbance in movement resulting from cerebellar dysfunction. It manifests as inaccurate movements with delayed onset and overshoot, especially when movements are repetitive or rhythmic. Identification of ataxia is integral to the diagnosis and assessment of severity, and is important in monitoring progression and improvement. Ataxia is identified and assessed by clinicians observing subjects perform standardised movement tasks that emphasise ataxic movements. Our aim in this paper was to use data recorded from motion sensors worn while subjects performed these tasks, in order to make an objective assessment of ataxia that accurately modelled the clinical assessment.MethodsInertial measurement units and a Kinect system were used to record motion data while control and ataxic subjects performed four instrumented version of upper extremities tests, i.e. Finger Chase Test (FCT), Finger Tapping Test (FTT), Finger to Nose Test (FNT) and Dysdiadochokinesia Test (DDKT). Kinematic features were extracted from this data and correlated with clinical ratings of severity of ataxia using the Scale for the Assessment and Rating of Ataxia (SARA). These features were refined using Feed Backward feature Elimination (the best performing method of four). Using several different learning models, including Linear Discrimination, Quadratic Discrimination Analysis, Support Vector Machine and K-Nearest Neighbour these extracted features were used to accurately discriminate between ataxics and control subjects. Leave-One-Out cross validation estimated the generalised performance of the diagnostic model as well as the severity predicting regression model.ResultsThe selected model accurately (96.4%) predicted the clinical scores for ataxia and correlated well with clinical scores of the severity of ataxia (rho = 0.8, p < 0.001). The severity estimation was also considered in a 4-level scale to provide a rating that is familiar to the current clinically-used rating of upper limb impairments. The combination of FCT and FTT performed as well as all four test combined in predicting the presence and severity of ataxia.ConclusionIndividual bedside tests can be emulated using features derived from sensors worn while bedside tests of cerebellar ataxia were being performed. Each test emphasises different aspects of stability, timing, accuracy and rhythmicity of movements. Using the current models it is possible to model the clinician in identifying ataxia and assessing severity but also to identify those test which provide the optimum set of data.Trial registrationHuman Research and Ethics Committee, Royal Victorian Eye and Ear Hospital, East Melbourne, Australia (HREC Reference Number: 11/994H/16).

show abstract