Filter Bank Regularized Common Spatial Pattern Ensemble for Small Sample Motor Imagery Classification

Area of Science:

• Neuroscience and Signal Processing
• Biomedical Engineering
• Machine Learning for Healthcare

Background:

• Motor imagery electroencephalograms (EEG) are valuable for brain-computer interfaces due to their non-invasiveness and high temporal resolution.
• Traditional Common Spatial Pattern (CSP) algorithms face performance degradation with limited data (small-sample setting) and require manual frequency band selection.
• Existing methods like filter bank CSP (FBCSP) offer improvements but can be further optimized.

Purpose of the Study:

• To introduce a novel feature extraction method for motor imagery EEG that overcomes the limitations of traditional CSP algorithms.
• To enhance classification accuracy, particularly in small-sample scenarios.
• To automate frequency band selection for improved efficiency and subject-specific adaptation.

Main Methods:

• The proposed method utilizes a filter bank to segment motor imagery EEG data.
• Regularized Common Spatial Pattern (R-CSP) is applied to the segmented data.
• Feature selection is performed using mutual information, followed by parameter set selection for an ensemble classifier.
• An ensemble classifier is employed for final classification based on the selected features.

Main Results:

• The proposed filter bank-based method demonstrated significant improvements in mean classification accuracy compared to CSP, SR-CSP, R-CSP, FBCSP, and SR-FBCSP.
• Accuracy gains ranged from 4.47% to 12.34% over existing methods.
• A notable improvement of 3.49% was observed compared to a parameter-selected version of filter bank R-CSP.
• The method showed particularly large performance gains in small-sample settings.

Conclusions:

• The proposed filter bank feature extraction method effectively addresses the limitations of traditional CSP, especially in small-sample scenarios.
• This approach offers a more robust and accurate solution for motor imagery EEG analysis.
• The automated frequency band selection and ensemble classification contribute to improved brain-computer interface performance.