A Deep Learning Framework for Assessing Physical Rehabilitation Exercises

Computer-aided assessment of physical rehabilitation entails evaluation ofpatient performance in completing prescribed rehabilitation exercises, based onprocessing movement data captured with a sensory system. Despite the essentialrole of rehabilitation assessment toward improved patient outcomes and reducedhealthcare costs, existing approaches lack versatility, robustness, andpractical relevance. In this paper, we propose a deep learning-based frameworkfor automated assessment of the quality of physical rehabilitation exercises.The main components of the framework are metrics for quantifying movementperformance, scoring functions for mapping the performance metrics intonumerical scores of movement quality, and deep neural network models forgenerating quality scores of input movements via supervised learning. Theproposed performance metric is defined based on the log-likelihood of aGaussian mixture model, and encodes low-dimensional data representationobtained with a deep autoencoder network. The proposed deep spatio-temporalneural network arranges data into temporal pyramids, and exploits the spatialcharacteristics of human movements by using sub-networks to process jointdisplacements of individual body parts. The presented framework is validatedusing a dataset of ten rehabilitation exercises. The significance of this workis that it is the first that implements deep neural networks for assessment ofrehabilitation performance.