Surface myoelectric signal analysis: dynamic approaches for change detection and classification

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Technology

Background:

• Surface myoelectric signals are crucial for controlling prosthetic limbs.
• Accurate detection and classification of these signals are essential for intuitive prosthesis function.
• Existing methods face challenges in real-time performance and classification accuracy.

Purpose of the Study:

• To develop a dynamic method for simultaneous detection and classification of events in surface myoelectric signals.
• To enhance the control capabilities of elbow and wrist prostheses.
• To compare the performance of novel classification techniques against established methods.

Main Methods:

• Utilized dynamic cumulative sum of local generalized likelihood ratios with wavelet decomposition for on-line event detection.
• Employed multiresolution wavelet analysis and autoregressive (AR) modeling for feature extraction.
• Applied polynomial classifiers for pattern modeling and matching, comparing them with neural networks and support vector machines.

Main Results:

• Achieved an average of 91% accuracy in detecting and classifying four elbow and wrist movements using wavelet features.
• Reached 95% accuracy with AR modeling features.
• Demonstrated superior classification accuracy and consistency with polynomial classifiers compared to neural networks and comparable results to support vector machines without parameter tuning.

Conclusions:

• The proposed dynamic method effectively detects and classifies myoelectric signal events for prosthetic control.
• Polynomial classifiers offer a robust and efficient alternative to neural networks and SVMs for this application.
• Further optimization may be needed to balance accuracy with desired short prosthesis response delays.