Adaptive HD-sEMG decomposition: towards robust real-time decoding of neural drive

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Neural interfacing relies on decoding electromyography (EMG) signals.
• High-density surface EMG (HD-sEMG) offers rich signal information but is susceptible to non-stationarities.
• Changes in joint pose and muscle contraction intensity challenge signal stability.

Purpose of the Study:

• To develop and validate an adaptive real-time algorithm for motor unit decoding from HD-sEMG.
• To assess the algorithm's robustness against signal non-stationarities.
• To compare the adaptive algorithm's performance against static decoding methods.

Main Methods:

• An adaptive real-time motor unit decoding algorithm was developed.
• The algorithm was tested on HD-sEMG data from the extensor carpi radialis brevis.
• Performance was benchmarked against intramuscular EMG decompositions during isometric contractions with varying wrist angles and intensities.

Main Results:

• The adaptive decoding algorithm demonstrated significantly higher decoding accuracies.
• Performance was superior in trials with differing contraction conditions compared to initialization data.
• Static decoding methods showed lower accuracy under non-stationary conditions.

Conclusions:

• Adaptive parameter tuning is crucial for robust neural decoding from HD-sEMG.
• Filter re-use decoding methods have limitations in non-stationary environments.
• The developed adaptive algorithm offers a more reliable approach for neural interfacing.