A Dimensionality Reduction Approach for Motor Imagery Brain-Computer Interface Using Functional Clustering and Graph...

Area of Science:

• Neuroscience
• Computer Science
• Signal Processing

Background:

• Motor imagery brain-computer interfaces (MI-BCI) rely on accurate electroencephalogram (EEG) signal classification.
• Dimensionality reduction is crucial for improving the efficiency and performance of MI-BCI systems.

Purpose of the Study:

• To introduce a novel framework for dimensionality reduction and classification of EEG signals in MI-BCI.
• To enhance the performance of MI-BCI systems, particularly in scenarios with limited training data.

Main Methods:

• The proposed Kron-reduced generic learning regularization with differential evolution (K-GLR-DE) framework integrates graph signal processing (GSP) and a meta-heuristic optimizer.
• Brain graphs are constructed and dimensionality reduction is achieved using physiological regions of interest (ROIs) and Kron reduction.
• Feature extraction involves graph total variation and generic learning regularized common spatial patterns (GLRCSP), followed by differential evolution (DE)-based feature selection.

Main Results:

• The K-GLR-DE approach was evaluated on the BCI Competition III Dataset IVa and the PhysioNet eegmmidb dataset.
• A support vector machine with a radial basis function (SVM-RBF) classifier achieved a mean accuracy of 96.46% ± 0.81% on BCIC III-IVa.
• The method demonstrated superior performance across various training conditions, including small and limited training sets.

Conclusions:

• The K-GLR-DE method significantly improves MI-BCI classification performance.
• The framework is effective even with limited training data, offering a robust solution for MI-BCI systems.