A Multimodal Biomedical Sensing Approach for Muscle Activation Onset Detection.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Muscle onset detection is crucial for electromyography (EMG) analysis in human-machine interaction and rehabilitation.
• Slow muscle activation processes present challenges due to low amplitude, long duration, and noise susceptibility.
• Accurate onset timing is vital for effective rehabilitation assessment and control.

Purpose of the Study:

• To propose and validate a lightweight temporal attention method for detecting slow muscle activation onset.
• To enhance the accuracy and robustness of onset detection in surface EMG signals.
• To address the challenges posed by slow activation processes and noise interference.

Main Methods:

• Developed a lightweight temporal attention framework for EMG signal analysis.
• Incorporated optical motion data for multimodal validation.
• Utilized temporal feature encoding, attention mechanism, and noise suppression strategies.
• Employed five-fold cross-validation on a diverse dataset of slow activation movements.

Main Results:

• The proposed method significantly outperformed traditional and deep learning baselines in accuracy, recall, and precision.
• Achieved ~92% accuracy, ~90% recall, and ~93% precision under normal conditions.
• Reduced average onset detection error to ~41ms and delay to ~28ms, with a ~2.2% false positive rate.
• Demonstrated robust and stable performance across varying noise levels and subjects.

Conclusions:

• The lightweight temporal attention method offers a highly accurate and robust solution for slow muscle activation onset detection.
• The approach is suitable for real-world applications in rehabilitation and human-machine interaction.
• The method shows strong generalization capabilities, making it reliable for diverse users and conditions.