Decoding Muscle Force From Motor Unit Firings Using Encoder-Decoder Networks.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Accurate interpretation of motor unit (MU) activities from surface electromyography (sEMG) is crucial for decoding motor intentions.
• Challenges include cross-trial MU tracking and information loss from incomplete decomposition.

Purpose of the Study:

• To present a novel framework for interpreting MU activities and decoding muscle force.
• To address limitations in current sEMG decomposition and MU tracking methods.

Main Methods:

• Clustering and classifying MUs based on spatially distributed firing waveforms for general MU tracking.
• Utilizing a deep network to predict normalized muscle force from MU firing trains and a twitch force model.
• Calibrating force levels using MU category distribution and compensating for missing MU data.

Main Results:

• The proposed framework achieved the lowest root mean square deviation (6.68% ± 1.29%) and highest fitness (R² of 0.94 ± 0.04) in muscle force estimation.
• Outperformed three common existing methods in predicting thumb abduction force from high-density sEMG data.

Conclusions:

• The developed framework offers a valuable computational solution for interpreting individual MU activities.
• Demonstrated effectiveness in accurate muscle force estimation, advancing noninvasive motor control research.