Subject-based feature extraction by using fisher WPD-CSP in brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalogram (EEG) feature extraction is crucial for brain-computer interfaces (BCIs).
• Common Spatial Pattern (CSP) is effective but requires many channels and lacks frequency information.
• Wavelet Packet Decomposition (WPD) combined with CSP addresses some limitations but overlooks subject-specific features.

Purpose of the Study:

• To propose a subject-based feature extraction method using Fisher WPD-CSP for improved BCI performance.
• To adapt the Fisher WPD-CSP algorithm to individual subjects for enhanced EEG analysis.
• To overcome the limitations of traditional CSP and WPD-CSP methods in capturing subject-specific neural patterns.

Main Methods:

• EEG signals decomposed into sub-bands using WPD.
• Average power values computed, and sub-bands with higher Fisher distance selected per subject.
• Selected sub-bands reconstructed as new channels for CSP input.
• A six-dimensional feature vector generated using CSP, followed by classification with a Probabilistic Neural Network (PNN).

Main Results:

• The subject-based Fisher WPD-CSP method demonstrated superior performance compared to non-subject-based Fisher WPD-CSP and WPD-CSP.
• Achieved high sensitivity (88.7±0.9%) and specificity (91±1%).
• Classification accuracy increased by 6-12% and 14% over the other methods, respectively.

Conclusions:

• The subject-based Fisher WPD-CSP method effectively addresses CSP's limitations by incorporating WPD and subject-specific feature selection.
• Fisher distance enhances the separability of selected sub-bands for individual subjects.
• This approach leads to significantly higher classification accuracy in BCI applications compared to existing WPD-CSP methods.