Rehabilitation Exergames: use of motion sensing and machine learning to quantify exercise performance in healthy volunteers

Background: Performing physiotherapy exercises in front of a physiotherapist yields qualitative assessment notes and immediate feedback. However, practicing the exercises at home lacks feedback on how well or not patients are performing the prescribed tasks. The absence of proper feedback might result in patients doing the exercises incorrectly, which could worsen their condition. Objective: We propose the use of two machine learning algorithms, namely Dynamic Time Warping (DTW) and Hidden Markov Model (HMM), to quantitively assess the patient’s performance with respects to a reference. Methods: Movement data were recorded using a Kinect depth sensor, capable of detecting 25 joints in the human skeleton model, and were compared to those of a reference. 16 participants were recruited to perform four different exercises: shoulder abduction, hip abduction, lunge, and sit-to-stand. Their performance was compared to that of a physiotherapist as a reference. Results: Both algorithms show a similar trend in assessing participants' performance. However, their sensitivity level was different. While DTW was more sensitive to small changes, HMM captured a general view of the performance, being less sensitive to the details. Conclusions: The chosen algorithms demonstrated their capacity to objectively assess physical therapy performances. HMM may be more suitable in the early stages of a physiotherapy program to capture and report general performance, whilst DTW could be used later on to focus on the detail