Anomaly Detection of Electromyographic Signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electromyographic (EMG) signal analysis is crucial for understanding neuromuscular activity.
• Detecting anomalies and contamination in EMG data is essential for accurate diagnosis and research.
• Existing methods may struggle with the complexity and variability of EMG signals.

Purpose of the Study:

• To develop a robust framework for detecting anomalous EMG signals.
• To identify different types of contamination within EMG datasets.
• To improve the reliability of EMG signal analysis through unsupervised anomaly detection.

Main Methods:

• Optimal Lawton wavelet transform for feature extraction.
• Robust Principal Component Analysis (rPCA) for dimensionality reduction.
• Self-Organizing Maps (SOM) and hierarchical clustering for anomaly separation and contamination identification.

Main Results:

• The framework successfully detected anomalous EMG signals in both synthetic and real-world datasets.
• High precision (99% ± 0.4) and recall (90% ± 3.3) were achieved in anomaly detection.
• The method effectively separated anomalous signals into distinct clusters for contamination identification.

Conclusions:

• The proposed framework offers a robust and effective solution for unsupervised anomaly detection in EMG signals.
• This methodology enhances the quality control of EMG data, leading to more reliable research outcomes.
• The approach demonstrates significant potential for real-time EMG signal quality assessment.