Improved artefact removal from EEG using Canonical Correlation Analysis and spectral slope

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Muscle artifacts are a common issue in electroencephalography (EEG) recordings, contaminating signals even at rest.
• These artifacts add significant high-frequency energy, hindering the interpretation of subcortical brain activity.
• Accurate detection and removal of muscle contamination are crucial for reliable EEG analysis.

Purpose of the Study:

• To introduce a novel automatic method for removing muscle artifacts from EEG data.
• To improve the accuracy and efficiency of artifact removal compared to existing techniques.
• To preserve underlying brain signals while effectively eliminating muscle contamination.

Main Methods:

• Developed an automatic muscle-removal approach combining Blind Source Separation-Canonical Correlation Analysis (BSS-CCA) with spectral slope analysis.
• Utilized the spectral slope of BSS-CCA components to discriminate between brain signals and muscle artifacts.
• Validated the method on a paralysis dataset to assess its performance.

Main Results:

• The proposed method effectively reduces high-frequency muscle contamination, particularly at peripheral EEG channels.
• Steady-state brain responses during cognitive tasks were preserved.
• The approach demonstrated improved performance over traditional BSS-CCA by addressing component mixing.

Conclusions:

• The enhanced BSS-CCA method with spectral slope rejection offers an automatic and effective solution for EEG muscle artifact removal.
• Performance is comparable to Independent Component Analysis (ICA) but with significantly lower computational complexity.
• This technique facilitates more reliable analysis of brain activity by reducing noise.