Canonical correlation analysis applied to remove muscle artifacts from the electroencephalogram

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalogram (EEG) signals are frequently obscured by muscle artifacts, compromising data quality.
• Existing methods for artifact removal, such as filtering or independent component analysis (ICA), have limitations.

Purpose of the Study:

• To develop and evaluate a new method for removing muscle artifacts from EEG data.
• To compare the efficacy of the proposed method against traditional techniques like low-pass filtering and ICA.

Main Methods:

• A novel muscle artifact removal technique based on canonical correlation analysis (CCA), a blind source separation (BSS) method, was developed.
• The method was tested using both synthetic EEG datasets and a real-world ictal EEG recording.
• Performance was evaluated against low-pass filtering and ICA-based artifact removal.

Main Results:

• The CCA-based method demonstrated superior performance in removing muscle artifacts compared to low-pass filtering and ICA.
• Artifact removal was successful on a real ictal EEG recording.
• The proposed method preserved the underlying ictal activity without significant alteration.

Conclusions:

• Canonical correlation analysis (CCA) provides an effective and robust approach for muscle artifact removal in EEG.
• This new method offers an improvement over existing techniques, ensuring cleaner EEG data for analysis.
• The successful application on real ictal EEG highlights its clinical relevance.