Detecting compensatory movements of stroke survivors using pressure distribution data and machine learning algorithms
Siqi Cai1, Guofeng Li1, Xiaoya Zhang2
1Shien-Ming Wu School of Intelligent Engineering, South China University of Technology, Guangzhou, 510640, China.
Journal of Neuroengineering and Rehabilitation
|November 6, 2019
Summary
This study developed a practical pressure distribution system to detect compensatory movements in stroke survivors during reaching tasks. Machine learning accurately identified these movements, aiding in rehabilitation and improving arm function.
Area of Science:
- Rehabilitation engineering
- Biomechanics
- Machine learning in healthcare
Background:
- Compensatory movements in stroke survivors can hinder long-term recovery of paretic arm function.
- Existing sensor and camera systems for detecting compensatory movements have limitations like obstruction and privacy concerns.
- A novel, unobtrusive system using pressure distribution data is proposed to overcome these limitations.
Purpose of the Study:
- To develop and validate a pressure distribution-based system for automatic detection of compensatory movements in stroke survivors.
- To apply machine learning algorithms for classifying compensatory movements during seated reaching tasks.
- To provide a practical and unobtrusive method for monitoring patient recovery.
Main Methods:
- Eight stroke survivors performed reaching tasks (back-and-forth, side-to-side, up-and-down) with both limbs.
- Pressure distribution data were collected, and five features were extracted for classification.
- K-nearest neighbor (k-NN) and support vector machine (SVM) algorithms classified movements, with surface electromyography (sEMG) used for detailed analysis.
Main Results:
- High classification accuracies (F1-score > 0.95) were achieved for both k-NN and SVM classifiers in detecting compensatory movements.
- Excellent discrimination between compensation and non-compensation (NC) movements (average F1-score of 0.993).
- Accurate multiclass classification of compensatory movement patterns (NC, trunk lean-forward, trunk rotation, shoulder elevation) with an average F1-score of 0.981.
Conclusions:
- The pressure distribution-based system demonstrated feasibility and effectiveness in detecting and categorizing compensatory movements.
- High classification accuracy by machine learning algorithms suggests potential for automatic monitoring of stroke survivors' compensatory movements during reaching tasks.
- This system offers a promising, unobtrusive approach to support stroke rehabilitation and improve motor function.
