Development and pilot test of a machine learning-based knee exercise system with video demonstration, real-time feedback, and exercise performance score

Abstract

Exercise therapy is a common and effective approach for managing chronic knee pain. However, individuals often receive minimal supervision from physical therapists when exercises are performed at home. In this study, we developed a video-based training system to allow individuals to perform lower limb exercises, based on a machine learning algorithm for pose detection and estimation. The system included three key features: (1) an exercise video demonstration, (2) real-time tracking and feedback of exercise movements, and (3) an overall score of exercise performance. We also pilot tested the system by having participants (n = 8) to use the system to perform lower limb exercises for 3 consecutive days. The results indicated that, compared with the baseline, the perceived usefulness of the system ( t = 3.25, p = 0.01) and perceived lower limb muscle strength ( t = 2.94, p = 0.02) significantly improved after 3 days. These findings provide knowledge about the initial views on this system by the participants. However, further enhancements of the features and full-scale experiments to examine the usability and acceptance of the system and its impact on knee health are needed.