Multi-domain feature joint optimization based on multi-view learning for improving the EEG decoding.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) systems are crucial for neurorehabilitation.
• Common Spatial Pattern (CSP) is a popular feature extraction technique for motor imagery (MI) classification.
• CSP's effectiveness depends on frequency band, time window, and channel selection in electroencephalogram (EEG) data.

Purpose of the Study:

• To propose a multi-domain feature joint optimization (MDFJO) method for enhanced MI classification.
• To improve the discriminative feature selection for BCI systems.
• To enhance the overall classification performance of BCI.

Main Methods:

• Utilized multi-view learning for optimal feature selection.
• Employed Fisher discriminant criterion (FDC) for channel pattern division.
• Applied CSP feature extraction on segmented EEG data across multiple sub-bands and time intervals.
• Introduced a feature sparsification strategy for temporal refinement.

Main Results:

• The MDFJO method achieved average classification accuracies of 88.29% (Data 1) and 87.21% (Data 2).
• MDFJO demonstrated significantly superior performance compared to MSO, FBCSP32, and other competing methods (p < 0.05).
• The proposed feature sparsification strategy effectively enhanced classification accuracy.

Conclusions:

• The MDFJO method significantly improves BCI test accuracy compared to CSP, SFBCSP, FBCSP, and MSO.
• The feature sparsification strategy is effective in enhancing classification accuracy.
• The proposed MDFJO method enhances the practicability and effectiveness of BCI systems.