An Automatic Channel Selection Approach for ICA-Based Motor Imagery Brain Computer Interface

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery brain-computer interfaces (MIBCI) utilize electroencephalogram (EEG) signals.
• Independent Component Analysis (ICA) is a spatial filtering technique for MIBCI.
• ICA-MIBCI performance is sensitive to EEG channel selection and data quality.

Purpose of the Study:

• To investigate EEG channel optimization for improving ICA-MIBCI stability and classification accuracy.
• To develop and validate a data-driven approach for selecting optimal EEG channels.
• To compare the performance of optimized ICA-MIBCI against standard methods.

Main Methods:

• A single-trial ICA-MIBCI system was developed using selected EEG channels.
• A 'self-testing' method was employed to evaluate single-trial data quality and select high-quality trials.
• Candidate channel configurations were tested, and optimal channels were selected based on self-testing accuracy.
• Common Spatial Pattern-based MIBCI (CSP-MIBCI) was used as a reference.

Main Results:

• Optimized channels improved ICA-MIBCI classification accuracy by 2.8% (self-testing), 14.4% (session-to-session), and 36.2% (subject-to-subject) compared to fixed channels.
• Compared to CSP-MIBCI, optimized channels showed relative accuracy increments of 8.5% (self-testing), 9.5% (session-to-session), and 26.7% (subject-to-subject).
• The proposed channel optimization method demonstrated significant performance gains across diverse transfer conditions.

Conclusions:

• EEG channel optimization is crucial for robust ICA-MIBCI performance.
• The 'self-testing' data quality evaluation and subsequent channel selection effectively enhance ICA-MIBCI.
• The proposed methodology offers a practical approach to improve the feasibility and reliability of ICA-MIBCI systems.