Functional Connectivity and Feature Fusion Enhance Multiclass Motor-Imagery Brain-Computer Interface Performance.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor-imagery brain-computer interfaces (MI-BCIs) face challenges in extracting discriminative features for multiple tasks.
• Traditional methods using single electroencephalography (EEG) channels neglect interconnections, limiting results.
• Functional brain connectivity (FC) shows promise but is hampered by high subject variability.

Purpose of the Study:

• To develop a novel signal processing framework to enhance feature extraction for multiclass MI-BCIs.
• To address the limitations of single-channel features and the variability issues with functional connectivity.
• To improve the performance and accuracy of MI-BCI systems.

Main Methods:

• Extracted translation-invariant features (TIFs) using a scattering convolution network (SCN).
• Extracted brain connectivity features (BCFs) to capture inter-channel relationships.
• Employed a feature fusion strategy combining TIFs and BCFs from selected channels.
• Utilized a multiclass support vector machine (SVM) for classification.

Main Results:

• The proposed feature fusion framework outperformed existing state-of-the-art methods on a public dataset (BCI Competition IV, dataset IIa).
• Merging TIFs with BCFs yielded superior results compared to using TIFs alone.
• Demonstrated the effectiveness of the combined feature approach in enhancing MI-BCI performance.

Conclusions:

• The developed framework offers a promising solution for improving multiclass MI-BCI system performance.
• Feature fusion of TIFs and BCFs is a key strategy for overcoming previous limitations.
• This approach has the potential to advance the field of brain-computer interfaces.