Correlation-Filter-Based Channel and Feature Selection Framework for Hybrid EEG-fNIRS BCI Applications

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-Computer Interfaces (BCI) integrate brain signals with external devices.
• Hybrid BCIs combining electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) offer complementary data.
• Effective feature and channel selection is crucial for optimizing hybrid BCI performance.

Purpose of the Study:

• To develop and validate a correlation-filter-based strategy for feature and channel selection in hybrid EEG-fNIRS BCIs.
• To fuse complementary information from EEG and fNIRS for enhanced classification accuracy.
• To identify the most effective filters and classifiers for motor imagery tasks in hybrid BCIs.

Main Methods:

• A correlation-based connectivity matrix was used to extract relevant EEG and fNIRS channels.
• Statistical features (slope, skewness, mean, kurtosis) were extracted and fused.
• Multiple filters (ReliefF, mRMR, chi-square, ANOVA, Kruskal-Wallis) were applied for feature selection.
• Traditional classifiers including neural networks, SVM, LDA, and ensembles were employed.

Main Results:

• The proposed correlation-filter framework significantly improved classification accuracy in hybrid EEG-fNIRS BCIs.
• The ReliefF filter combined with an ensemble classifier achieved the highest accuracy of 94.77 ± 4.26%.
• Statistical analysis confirmed the significance of the results (p < 0.01).

Conclusions:

• The developed correlation-filter-based channel and feature selection strategy is effective for hybrid EEG-fNIRS BCIs.
• This approach enhances classification performance, paving the way for more robust BCI applications.
• The findings suggest the potential for widespread adoption in future hybrid BCI systems.