KCS-FCnet: Kernel Cross-Spectral Functional Connectivity Network for EEG-Based Motor Imagery Classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI) tasks are crucial for brain-computer interfaces (BCIs).
• Classifying hand laterality in MI tasks using electroencephalography (EEG) presents challenges in feature extraction and interpretability.
• Existing methods often lack rich spatial-temporal-spectral feature representation.

Purpose of the Study:

• To introduce a novel method, Kernel Cross-Spectral Functional Connectivity Network (KCS-FCnet), for improved EEG-based MI classification.
• To enhance the interpretability and efficiency of EEG-driven MI discrimination.
• To develop a more adaptable approach for individual subject characteristics in MI tasks.

Main Methods:

• Utilized a single 1D-convolutional neural network to extract temporal-frequency features from raw EEG data.
• Employed a cross-spectral Gaussian kernel connectivity layer to model functional relationships between EEG channels.
• Developed Kernel Cross-Spectral Functional Connectivity Network (KCS-FCnet) for spatial-temporal-spectral feature mapping.

Main Results:

• KCS-FCnet provides richer spatial-temporal-spectral feature maps compared to existing methods.
• The functional connectivity feature map significantly reduces the number of parameters, enhancing model interpretability.
• Demonstrated effective classification of right and left-hand classes in MI tasks with adaptable patterns for individual subjects.

Conclusions:

• KCS-FCnet is a promising shallow architecture for EEG-based MI classification.
• The method offers improved interpretability and efficiency for EEG-driven MI discrimination.
• KCS-FCnet shows potential for real-world applications in brain-computer interface systems.