Limitations and applications of ICA for surface electromyogram

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Surface electromyogram (SEMG) recordings are crucial for various applications but are often corrupted by artifacts and noise, especially during low muscle activity.
• Separating SEMG signals from noise and other bioelectric signals is challenging due to spectral and temporal overlap.

Purpose of the Study:

• To evaluate the efficacy of Independent Component Analysis (ICA) for separating muscle activity and removing artifacts from SEMG signals.
• To investigate factors influencing the reliability of ICA-based separation, including signal properties and the number of sources.
• To propose and test a robust measure for assessing the quality of bioelectric signal separation.

Main Methods:

• Application of Independent Component Analysis (ICA) to SEMG data.
• Evaluation of signal properties and the number of independent sources.
• Testing Zibulevsky's temporal plotting technique for source identification.
• Development and validation of a novel error measure using simulated and estimated mixing matrices.

Main Results:

• ICA is suitable for SEMG signal separation, even at low muscle activity levels.
• ICA's performance degrades when the number of sources exceeds the number of recording channels.
• Limitations in identifying the correct order and magnitude of separated signals were observed.
• Zibulevsky's technique proved insufficient for identifying independent sources even after filtering, due to insufficient data sparsity.

Conclusions:

• ICA is a viable tool for SEMG artifact removal and muscle activity separation.
• The number of recording channels relative to the number of signal sources is a critical factor for successful ICA application.
• A robust error measure is essential for quantifying the quality of ICA-based signal separation in bioelectric applications.