A novel approach to tele-rehabilitation: Implementing a biofeedback system using machine learning algorithms

Abstract

Tele-rehabilitation (Tele-rehab) is changing the landscape of virtual care by redefining assessment and breaking accessibility barriers as a convenient substitute for conventional rehabilitation. The COVID-19 pandemic resulted in a rapid uptake of virtual care. Researchers and health professionals have started developing new tele-rehab platforms, e.g., in the form of video conferencing. Albeit useful, these platforms still require the clinicians’ time and energy. Integrating a biofeedback system that can reliably distinguish between “Correctly Executed” from “Incorrectly Executed” exercises into tele-rehab platforms can help patients to perform rehab exercises correctly, avoid injuries, and enhance recovery. To address this gap, this paper proposes an automated system that uses machine learning to classify correct and incorrect executions of 9 rehabilitation gestures. The model is trained on 24 angle signals extracted from different body sections. The angle signals are obtained in 3D space, and 10 features are extracted from each signal. Six different classifiers, including Random Forest, Multi-Layer Perceptron Artificial Neural Networks, Naïve Bayes, Support Vector Machine, K-Nearest Neighbors, and Logistic Regression, are used, and evaluated with 10-Fold and Leave One Subject Out (LOSO) cross validations. The best classifiers achieved an average accuracy of 89.86% ± 3.38% and F1-Score of 72.84% ± 11.98% for 10-Fold and an average accuracy of 88.21% ± 3.90% and F1-Score of 68.16%±13.28% for LOSO. The proposed system has great potential to be integrated into tele-rehab platforms to help patients perform their exercises reliably.© 2017 Elsevier Inc. All rights reserved.