An Integrated System for Stroke Rehabilitation Exercise Assessment Using KINECT v2 and Machine Learning

Abstract

Stroke-induced physical disabilities necessitate consistent and effective rehabilitation exercises. While a typical regime encompasses 20–60 min daily, ensuring adherence and effectiveness remains a challenge due to lengthy recovery periods, potential demotivation, and the need for professional supervision. This paper presents an innovative home-based rehabilitation system designed to address these challenges by leveraging the capabilities of the KINECT v2 3D camera. Our system, equipped with a graphical user interface (GUI), allows patients to perform, monitor, and record their exercises. By utilizing advanced machine learning algorithms, specifically G3D and disentangled multi-scale aggregation schemes, the system can analyze exercises, generating both primary objective (PO) and control factor (CF) scores out of 100. This scoring assesses the exercise quality, providing actionable feedback for improvement. Our model is trained on the Kinematic Assessment of Movement and Clinical Scores for Remote Monitoring of Physical Rehabilitation (KIMORE) dataset, ensuring robust real-time scoring. Beyond scoring, the system offers pose-correction recommendations, ensuring exercises align with expert guidelines. It can evaluate the efficacy of five distinct exercises, with provision for including more based on individual needs and expert recommendations. Overall, our system offers a streamlined approach to stroke rehabilitation, promising enhanced feasibility, and patient engagement, potentially revolutionizing stroke recovery in the healthcare domain.