Exploring the Feasibility of Computer Vision for Detecting Post-Stroke Compensatory Movements
Summary
This study developed a system to detect compensatory movements in stroke survivors using a tablet camera and MediaPipe. The technology achieved high accuracy, aiding stroke motor recovery.
Area of Science:
- Rehabilitation Technology
- Biomedical Engineering
- Neuroscience
Background:
- Compensatory movements are common after stroke and can hinder long-term motor recovery.
- Objective monitoring of movement quality is crucial for effective stroke rehabilitation.
Purpose of the Study:
- To develop and evaluate a system for detecting compensatory movements during seated reaching tasks in stroke survivors.
- To utilize a conventional tablet camera and markerless pose estimation for movement analysis.
Main Methods:
- Employed MediaPipe, an open-source markerless body pose tracking algorithm, with a tablet camera.
- Annotated compensatory movements by comparing paretic and non-paretic arm movements frame-by-frame.
- Trained a binary classification model using the XGBoost algorithm to identify compensatory movements.
Main Results:
- The XGBoost model achieved an average accuracy of 0.92 (SD 0.07) in detecting compensatory movements.
- Challenges included occasional missing landmarks and misalignment issues with the MediaPipe Pose algorithm.
- Demonstrated the feasibility of near real-time compensatory movement detection.
Conclusions:
- A camera-based system with markerless pose tracking shows promise for detecting compensatory movements in stroke rehabilitation.
- Further research is needed to improve generalizability across diverse stroke survivor populations and enable mobile device implementation.
