Automated classification of bimanual movements in stroke telerehabilitation: a comparison of dimensionality reduction algorithms

Stroke survivors commonly experience unilateral muscle weakness, which limits their engagement in daily activities. Bimanual training has been demonstrated to effectively recover coordinated movements among those patients. We developed a low cost telerehabilitation platform dedicated to bimanual exercise, where the patient manipulates a dowel to control a computer program. Data on movement is collected using a Microsoft Kinect sensor and an inertial measurement unit to interface the platform, as well as to assess motor performance remotely. Toward automatic classification of bimanual movements executed by the user, we test the performance of a linear and a nonlinear dimensionality reduction techniques.