Unsupervised learning technique for surface electromyogram denoising from power line interference and baseline...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Electrophysiology

Background:

• Surface electromyograms (EMG) are often corrupted by power line interference (PLI) and baseline wander (BW).
• Accurate removal of these artifacts is crucial for reliable EMG analysis in clinical and research settings.
• Existing methods may not effectively separate these specific artifacts from the underlying EMG signal.

Purpose of the Study:

• To develop and evaluate a novel method for removing single-channel PLI and BW from EMG signals.
• To leverage non-negative matrix factorization (NMF) with prior knowledge of interference characteristics.
• To enhance the performance of artifact removal compared to existing state-of-the-art techniques.

Main Methods:

• The proposed approach utilizes non-negative matrix factorization (NMF) on the EMG signal's spectrogram.
• It decomposes the spectrogram into non-negative components representing PLI, BW, and EMG.
• Initialization incorporates a priori knowledge of PLI, BW, and EMG time-frequency patterns and optimized decomposition rank.

Main Results:

• The method effectively separates PLI and BW, which are sparse, from the noise-like EMG signal.
• Initialization with specific signal structures and adjusted rank significantly improved separation performance.
• Comparative analysis demonstrated superior performance over two established reference methods.

Conclusions:

• The presented NMF-based approach offers an effective solution for PLI and BW removal in single-channel EMG.
• The method's reliance on signal characteristics and optimized NMF parameters enhances artifact separation.
• This technique holds promise for improving the quality of EMG data for various applications.