Computer-vision based method for quantifying rising from chair in Parkinson's disease patients

The ability to arise from a sitting standing position is often impaired in Parkinson's disease (PD). This impairment associated with an increased risk of falling, and higher dementia. We propose novel approach estimate Movement Disorder Society Unified PD Rating Scale (MDS-UPDRS) ratings for “item 3.9” (arising chair) using computer vision-based method, whereby we use clinically informed reasoning engineer small number informative features high dimensional markerless pose estimation data. analysed 447 videos collected via the KELVIN-PD™ platform, recorded clinical settings at multiple sites, commercially available mobile smart devices. Each video showed examination item 3.9 MDS-UPDRS had severity rating trained clinician on 5-point scale (0, 1, 2, 3 or 4). deep learning library OpenPose was used extract key points each frame videos, resulting time-series signals point. From these signals, were extracted which capture relevant characteristics movement; velocity variation, smoothness, whether patient their hands push themselves up, how stooped while upright when fully standing. These train ordinal classification system (with one class possible UPDRS), based series random forest classifiers. UPDRS estimated by this system, leave-one-out cross validation, corresponded exactly made clinicians 79% within those 100% cases. able distinguish normal Parkinsonian movement sensitivity 62.8% specificity 90.3%. Analysis misclassified examples highlighted potential detect potentially mislabelled show that our computer-vision method can accurately quantify patients’ perform arising chair action. As far as are aware first study estimating scores singular monocular video. help prevent human error identifying unusual ratings, provides promise such being routinely assessments, either locally remotely, stratification outcome measures trials.