A Deep Learning Framework for Efficient Online Decomposition of High-Density Surface Electromyogram into Motor Unit...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Accurate decomposition of high-density surface electromyogram (HD-sEMG) into motor unit spike trains (MUSTs) is crucial for understanding neural motor control.
• Existing online decomposition methods face challenges in efficiency and accuracy.

Purpose of the Study:

• To present a novel deep learning (DL) framework for fast and accurate online decomposition of HD-sEMG signals.
• To improve the identification and reconstruction of motor unit spike trains (MUSTs) from HD-sEMG data.

Main Methods:

• A hybrid DL model was developed to capture spatiotemporal features for MU spike identification.
• A post-processing approach was used for precise reconstruction of continuous MUSTs.
• An asymmetric loss function was integrated to handle imbalanced sample data.

Main Results:

• The proposed framework achieved online MUSTs estimation from HD-sEMG with low latency (~7.5 ms).
• High matching rates were obtained across simulated, experimental, and public datasets, outperforming baseline methods.
• The framework demonstrated significant accuracy in identifying motor unit activity.

Conclusions:

• The novel DL framework offers an efficient and accurate solution for online HD-sEMG decomposition.
• This technology has the potential to advance real-time applications in neurorehabilitation medicine and neural motor control.
• The study provides a valuable tool for researchers and clinicians working with neuromuscular signals.