Attention-Guided Deep Learning Framework for Movement Quality Assessment

Physical rehabilitation programs frequently begin with a brief stay in the hospital and continue with home-based rehabilitation. Lack of feedback on exercise correctness is a significant issue in home-based rehabilitation. Deep learning-based movement quality assessment (MQA) can assist with home-based rehabilitation by providing the necessary quantitative feedback. However, systems developed for home-based settings must be fast and offer interpretability of the generated assessment. In this paper, we explore an attention-guided transformer-based architecture for MQA. A comparative analysis against the current state-of-the-art methods is undertaken to establish the validity of the proposed model. Further, we show that the proposed model offers significant performance improvement in training and inference time, which is pivotal for any real-time system. Finally, we show that analysis of the attention maps of the proposed model can give critical insights into the decision-making process of the deep-learning model, thus improving the overall interpretability of predicted assessment scores.