User adaptation in long-term, open-loop myoelectric training: implications for EMG pattern recognition in prosthesis
Jiayuan He1, Dingguo Zhang, Ning Jiang
1State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai, People's Republic of China.
Journal of Neural Engineering
|June 2, 2015
Summary
Electromyographic (EMG) signal classification performance degrades over time, but improves with repeated use. Learning curves for EMG pattern recognition in prosthesis control show exponential adaptation in both able-bodied and amputee users.
Area of Science:
- Biomedical Engineering
- Rehabilitation Engineering
- Signal Processing
Background:
- Electromyographic (EMG) signal classification performance degrades over time without retraining.
- This degradation impacts the usability of EMG pattern recognition for active prosthesis control.
Purpose of the Study:
- Investigate changes in EMG classification performance over 11 consecutive days.
- Analyze user adaptation characteristics in myoelectric control.
Main Methods:
- Tested EMG classification performance across 11 days in 8 able-bodied subjects and 2 amputees.
- Trained classifiers on one day's data and tested on subsequent days.
Main Results:
- Classification error decreased exponentially and plateaued after 4 days (able-bodied) and 6-9 days (amputees).
- Between-day performance approached within-day performance over time.
- EMG signal features showed progressively smaller changes with increased task performance days.
Conclusions:
- Increased practice leads to more consistent motor task performance and repeatable EMG patterns.
- User adaptation in myoelectric control can be modeled by exponential learning curves.
- Findings inform the design of adaptive pattern recognition systems for long-term myoelectric control.
