Improving Long Term Myoelectric Decoding, Using an Adaptive Classifier with Label Correction.

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Neuroprosthetics

Background:

• Myoelectric decoding algorithms for prosthesis control suffer from accuracy degradation over time due to muscle signal variations.
• Existing algorithms struggle to adapt to both gradual and rapid changes in electromyography (EMG) signals.

Purpose of the Study:

• To develop a novel adaptive myoelectric decoding algorithm for upper limb prosthesis control.
• To address and mitigate the inherent decay in decoding accuracy observed in current myoelectric systems.

Main Methods:

• The proposed algorithm utilizes unsupervised, on-demand updates of the training set to adapt to signal changes.
• It incorporates training data updates for slow signal variations and label correction for fast signal variations.
• Myoelectric data was collected from an able-bodied user over 4.5 hours during eight distinct wrist movements.

Main Results:

• The adaptive algorithm demonstrated a significantly lower decay rate of 0.2 per hour compared to 3.3 per hour for a non-adaptive classifier.
• This indicates a substantial improvement in maintaining decoding accuracy over extended periods.

Conclusions:

• The developed adaptive algorithm effectively maintains long-term decoding accuracy in EMG signals.
• This advancement promises to enhance the overall performance and reliability of myoelectric upper limb prostheses.