A Non-Intrusive Neural Quality Assessment Model for Surface Electromyography Signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Surface electromyography (sEMG) measurements are often contaminated by electrocardiogram (ECG) signals, especially near the heart.
• Accurate assessment of sEMG data quality is crucial for reliable interpretation in clinical and research settings.

Purpose of the Study:

• To develop and validate QASE-net, a non-intrusive model for predicting the signal-to-noise ratio (SNR) of sEMG signals.
• To enhance the effectiveness of real-world sEMG data quality assessment.

Main Methods:

• QASE-net employs an end-to-end trained architecture integrating a 1D Convolutional Neural Network (CNN), a Bidirectional Long Short-Term Memory (BLSTM) layer, and attention mechanisms.
• Utilized real-world sEMG data from the Non-Invasive Adaptive Prosthetics Database and ECG data from the MIT-BIH Normal Sinus Rhythm Database for experimental validation.

Main Results:

• QASE-net demonstrated superior performance compared to a baseline method in predicting sEMG signal-to-noise ratio.
• The model exhibited significantly reduced prediction errors and higher linear correlations with ground truth SNR values.
• Experimental results confirmed the effectiveness of QASE-net in assessing sEMG data quality.

Conclusions:

• QASE-net offers a promising solution for improving the reliability and precision of sEMG quality assessment in practical scenarios.
• The proposed model has the potential to significantly enhance the usability of sEMG data in various applications.
• This work contributes to more robust signal processing techniques for biomedical measurements.