A Kernel-based Nonlinear Manifold Learning for EEG Channel Selection with Application to Alzheimer's Disease
This paper introduces a novel EEG channel selection method to determine which channel interrelationships provide the best classification accuracy between a group of patients with Alzheimer's disease (AD) and a cohort of age matched healthy controls (HC). Thereby, determine which inter-relationships are more important for the in-depth dynamical analysis to further understand how neurodegenerative diseases such as AD affects global and local brain dynamics. The channel selection methodology uses kernel-based nonlinear manifold learning via Isomap and Gaussian Process Latent Variable Model (Isomap-GPLVM). The Isomap-GPLVM method is employed to learn both the spatial and temporal local similarities and global dissimilarities present within the EEG data structures. The resulting kernel (dis)similarity matrix is used as a measure of synchrony between EEG channels. Based on this information, channel-specific linear Support Vector Machine (SVM) classification is then used to determine which spatio-temporal channel inter-relationships are more important for in-depth dynamical analysis. In this work, the analysis of EEG data from HC and AD patients is presented as a case study. Our analysis shows that inter-relationships between channels in the fronto-parietal region and the rest are better at differentiating between AD and HC groups.