An EEG channel selection method for motor imagery based on Fisher score and local optimization

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Multi-channel electroencephalogram (EEG) offers high spatial resolution in brain-computer interface (BCI) research.
• Increased channel count in EEG-BCI systems necessitates longer data processing, hindering rapid response.
• Efficient EEG channel selection is crucial for maintaining decoding effectiveness while reducing computational load.

Purpose of the Study:

• To develop and evaluate a novel within-subject EEG channel selection method for BCI applications.
• To reduce the number of EEG channels without compromising decoding accuracy.
• To enhance the portability and efficiency of BCI systems.

Main Methods:

• Proposed a local optimization method utilizing Fisher scores for EEG channel selection.
• Extracted common spatial pattern characteristics across different frequency bands.
• Ranked channels based on Fisher scores and applied local optimization for final selection.

Main Results:

• Achieved an average accuracy of 79.37% using 11 selected channels on the BCI Competition IV Dataset IIa, outperforming full 22-channel usage by 6.52%.
• Demonstrated a 24.20% accuracy improvement with fewer than half the channels on a self-collected dataset, reaching 76.95% average accuracy.
• Highlighted the significance of channel combinations in enhancing selection quality.

Conclusions:

• The proposed Fisher score-based local optimization method effectively selects fewer EEG channels with higher accuracy for BCI.
• Channel selection and combination strategies improve BCI system portability and performance.
• This approach holds potential for developing practical, portable BCI systems for motor imagery classification.