Kernel-based Nonlinear Manifold Learning for EEG-based Functional Connectivity Analysis and Channel Selection with...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Informatics

Background:

• Electroencephalogram (EEG) analysis is vital for understanding neurological disorders.
• Selecting key EEG channels reduces complexity and improves diagnostic accuracy.
• Functional connectivity (FC) measures are essential for EEG-based diagnostics.

Purpose of the Study:

• To develop a generic measure of (dis)similarity for EEG functional connectivity analysis and channel selection.
• To apply kernel-based nonlinear manifold learning for enhanced EEG channel selection.
• To investigate functional connectivity differences between Alzheimer's disease patients and healthy controls.

Main Methods:

• Utilized Isomap and Gaussian Process Latent Variable Model (Isomap-GPLVM) for learning (dis)similarity in EEG data.
• Developed a novel kernel (dis)similarity matrix as a measure of linear and nonlinear FC.
• Analyzed EEG data from healthy controls (HC) and Alzheimer's disease (AD) patients.

Main Results:

• Identified significant differences in functional connectivity between occipital and other brain regions in AD patients compared to HC.
• Demonstrated that FC changes along the fronto-parietal region are critical for AD diagnosis.
• Achieved classification results comparable to other common FC measures.

Conclusions:

• Kernel-based nonlinear manifold learning provides a robust method for EEG channel selection and FC analysis.
• The proposed method effectively distinguishes between AD patients and HC based on EEG functional connectivity.
• Findings align with previous neuroimaging studies, highlighting the utility of EEG in characterizing Alzheimer's disease.