Estimating the Quality of Reaching Movements in Stroke Survivors

Stroke is a leading cause of permanent impairments worldwide. Stroke survivors can improve their motor function through intensive and longitudinal rehabilitation therapies. However, rehabilitation therapy in outpatient settings often does not provide a sufficient amount of rehabilitation for meaningful patients’ recovery. Hence, patients need to perform therapeutic exercises in their home settings. However, despite the widely acknowledged importance of executing exercise movements in therapeutically desirable ways, stroke patients often practice exercise movements in a therapeutically undesirable manner, especially when therapists’ supervision is not available. While wearable sensors have been investigated to monitor patients’ exercise movements, it remains as a challenge to accurately and objectively estimate the quality of at-home exercise movements. This study proposes an analytic method to estimate the quality of individual reaching movements in stroke survivors using a wrist-worn wearable inertial sensor. We extracted time/frequency-domain features from stroke survivor’s horizontal movements in reaching and trained supervised machine learning models to classify if each reaching movement is executed in a therapeutically desirable way or not. The estimation results show that acceptable performance with the Area Under the Receiver Operating Characteristic Curve of 0.94. The extracted temporal and frequency features demonstrate that significant differences exist between the two classes and reflect the level of impairments in reaching movements. We envision these results enable monitoring the quality of at-home exercise movements and personalizing rehabilitation therapy.