[Three-class Motor Imagery Classification Based on Optimal Sub-band Features of Independent Components]

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCI) using scalp electroencephalogram (EEG) face challenges with individual differences and background noise, impacting system stability.
• Optimization of BCI system parameters, including temporal/spatial filters and classifier settings, is crucial for improving accuracy.
• Existing methods often struggle to account for inter-subject variability and complex signal interference.

Purpose of the Study:

• To propose and evaluate a novel BCI information processing method to enhance classification accuracy.
• To address the limitations of individual differences and background noise in EEG-based BCI systems.
• To investigate the efficacy of combining optimized independent component analysis spatial filters (ICA-SF) with multiple sub-band EEG features.

Main Methods:

• Developed a new BCI information processing method integrating ICA-SF optimization with multiple sub-band EEG features.
• Analyzed three-class motor imagery EEG (MI-EEG) data from four subjects collected over different periods.
• Performed inner/outer cross-validation within subjects and subject-to-subject validation.

Main Results:

• The proposed multiple sub-band method consistently achieved higher average classification accuracy compared to single-band methods across all validation schemes.
• Significant accuracy improvements were observed, with maximum differences of 6.08% in single-subject validation and 5.15% in subject-to-subject validation.
• The method demonstrated robustness in handling individual differences and improving BCI performance.

Conclusions:

• The proposed BCI information processing method effectively improves classification accuracy by leveraging multiple sub-band features and optimized ICA-SF.
• This approach offers a more stable and accurate solution for EEG-based BCI systems, particularly in the presence of individual variability.
• The findings suggest a promising direction for advancing motor imagery BCI applications.