Post-stroke Exercise Assessment using Hybrid Quantum Neural Network

Post-stroke patient rehabilitation involves physical therapy primarily to re-learn activities due to the damage caused by stroke. The patient performs these therapies in the in-patient clinical care center under the direct supervision of clinicians to monitor the progress. However, these therapy services are costly and limited availability due to the lack of clinicians, constraint dependent (e.g., clinic services only). With technological advancement and fast-growing artificial intelligence (AI), human movement can be captured through skeletal joint movement using various sensors (e.g., Kinect) and develop AI-enabled applications to assess the performed exercise automatically. In this work, we proposed a novel hybrid quantum neural network model to evaluate post-stroke patients’ exercises sensed through Kinect sensors. Our proposed hybrid quantum neural network model comprises traditional neural network layers and quantum layers to extract inherent hidden features uncaptured by neural networks and perform assessment tasks. We showcase the effectiveness and efficacy of our model by evaluating the performance using a publicly available dataset (UI-PRMD) that consists of ten exercises from 10 users. Our proposed model achieves RMSE error on average 0.044341 and outperforms the traditional machine learning algorithms.