JMIR Publications

ABSTRACT

Background:

Human interaction with machines is essential for the success of telerehabilitation programs. Telerehabilitation devices are designed for use by individuals whose behavior is atypical due to motor impairment. In order to implement optimal control strategies for human-machine interactions that are intuitive and safe, the machine must “gain an understanding” of the user's actions.

Objective:

This study seeks to demonstrate the possibility of classifying bimanual movements in telerehabilitation using machine learning, toward automatic assessment of motor performance.

Methods:

Nine healthy individuals interacted with a commercial virtual reality gaming system within an engaging citizen science project. Time series of their movement were recorded by the sensors embedded in the device as they performed scientific tasks. We performed principal component analysis to identify salient features of movements, and then applied a bagged trees ensemble classifier to classify the movements.

Results:

Classification achieved exceptionally high performance, reaching 99.9% accuracy. Among the movements, elbow flexion was most accurately classified (99.2%) and horizontal shoulder abduction to the right side of the body was most misclassified (98.8%).

Conclusions:

Coordinated bimanual movements in virtual reality can be classified with extraordinary high accuracy. Our findings lay the foundation for the development of motion analysis algorithms in virtual reality-mediated telerehabilitation.