A multi-label deep residual shrinkage network for high-density surface electromyography decomposition in real-time
Jinting Ma1, Lifen Wang1, Renxiang Wu1
1School of Biomedical Engineering, Shenzhen University Medical School, Shenzhen University, Shenzhen, Guangdong, China.
Journal of Neuroengineering and Rehabilitation
|May 9, 2025
Summary
A new algorithm, ML-DRSNet, significantly improves the accuracy and reduces latency in identifying motor unit spike trains (MUSTs) from surface electromyography (sEMG). This advancement is crucial for real-time neural interface control.
Area of Science:
- Biomedical Engineering
- Neuroscience
- Signal Processing
Background:
- Accurate identification of motor unit spike trains (MUSTs) from surface electromyography (sEMG) is critical for real-time neural interface control.
- Existing sEMG decomposition methods suffer from high latency or low accuracy, hindering practical applications.
Purpose of the Study:
- To introduce a novel real-time high-density sEMG (HD-sEMG) decomposition algorithm, ML-DRSNet.
- To enhance accuracy and reduce latency in sEMG decomposition for improved neural interface performance.
Main Methods:
- Developed ML-DRSNet, combining multi-label learning with a deep residual shrinkage network (DRSNet).
- Evaluated ML-DRSNet on a public sEMG dataset, comparing it with ML-DCNN and MT-DCNN.
- Tested algorithm performance using various window and step sizes.
Main Results:
- ML-DRSNet achieved significantly higher decomposition precision (0.86 ± 0.18) compared to ML-DCNN (0.71 ± 0.24) and MT-DCNN (0.66 ± 0.16).
- ML-DRSNet demonstrated substantially lower latency (15.15 ms) than ML-DCNN (69.36 ms) and MT-DCNN (76.96 ms).
Conclusions:
- ML-DRSNet and ML-DCNN algorithms significantly improve accuracy and real-time performance in MUST decomposition.
- These advancements provide a foundation for neuro-information-driven motor intention recognition and disease assessment.
