Robust filter bank common spatial pattern (RFBCSP) in motor-imagery-based brain-computer interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery-based Brain Computer Interfaces (MI-BCI) rely on electroencephalogram (EEG) signal analysis.
• The Filter Bank Common Spatial Pattern (FBCSP) algorithm is effective for feature extraction in MI-BCI.
• FBCSP's sensitivity to outliers in EEG data can degrade classification performance.

Purpose of the Study:

• To introduce a Robust FBCSP (RFBCSP) algorithm for improved MI-BCI performance.
• To enhance the robustness of EEG feature selection by addressing outlier sensitivity.
• To evaluate the effectiveness of RFBCSP compared to the standard FBCSP algorithm.

Main Methods:

• Developed the Robust FBCSP (RFBCSP) algorithm by integrating the Minimum Covariance Determinant (MCD) estimator.
• Replaced standard covariance matrix estimations in FBCSP with robust MCD estimates.
• Evaluated RFBCSP performance on a public EEG dataset using cross-validation and session-to-session transfer.

Main Results:

• RFBCSP demonstrated improved classification accuracies for certain subjects compared to FBCSP.
• A slight overall performance improvement was observed across subjects with RFBCSP.
• Analysis indicated that RFBCSP effectively excluded outliers during covariance matrix estimation.

Conclusions:

• RFBCSP offers a promising approach for robust EEG classification in MI-BCI.
• The integration of robust statistical methods enhances the reliability of BCI systems.
• Further research into RFBCSP could lead to more dependable brain-computer interfaces.