Online detection of compensatory strategies in human movement with supervised classification: a pilot study

Introduction Stroke survivors often compensate for the loss of motor function in their distal joints by altered use more proximal and body segments. Since this can be detrimental to rehabilitation process long-term, it is imperative that such movements are indicated patients caregiver. This a difficult task since compensation strategies varied multi-faceted. Recent works have focused on supervised machine learning methods detection require large training dataset motions with location annotations each time-step recorded motion. In contrast, study proposed novel approach learned linear classifier from energy-based features discriminate between healthy compensatory identify compensating without need dense explicit annotations. Methods Six physiotherapists performed five different tasks using acted compensations. The resulting motion capture data was transformed into joint kinematic dynamic trajectories. Inspired bio-mechanics, were extracted dataset. Support vector (SVM) logistic regression (LR) algorithms then applied movements. For identification, an additional condition enforcing independence feature calculation observable degree freedom imposed. Results Using leave-one-out cross validation, low values mean brier score (<0.15), mis-classification rate (<0.2) false discovery obtained both SVM LR classifiers. These found outperform deep classifiers did not features. Additionally, online classification performance our also shown baselines. Furthermore, qualitative results identification experiment method could successfully joints. Discussion including prior bio-mechanical information form energy based improve even when used, offline classification. evaluation algorithm potentially provide straightforward interpretable way identifying joints, as well being performed.