Background: The microstructural EEG elements and their functional networks relate to many neurophysiological functions of the brain and can reveal abnormalities. Despite the blooming variety of methods for estimating connectivity in the EEG of a single subject, a common pitfall is seen in relevant studies; grand averaging is used for estimating the characteristic connectivity patterns of a group of subjects. This averaging may distort results and fail to account for the internal variability of connectivity results across the subjects of a group.
New method: In this study, we propose a novel methodology for the cross-subject network investigation of EEG graphoelements. We used dimensionality reduction techniques in order to reveal internal connectivity properties and to examine how consistent these are across a number of subjects. In addition, graph theoretical measures were utilized to prioritize regions according to their network attributes.
Results: As proof of concept, we applied this method on fast sleep spindles across 10 healthy subjects. Neurophysiological findings revealed subnetworks of the spindle events across subjects, highlighting a predominance for occipito-parietal areas and their connectivity with frontal regions.
Comparison with existing methods: This is a new approach for the examination of within-group connectivities in EEG research. The results accounted for more than 85% of the overall data variance and the detected subnetworks were found to be meaningful down-projections of the grand average of the group, suggesting sufficient performance for the proposed methodology.
Conclusion: We conclude that the proposed methodology can serve as an observatory tool for the EEG connectivity patterns across subjects, providing a supplementary analysis of the existing topography techniques.
Keywords: EEG microstructure; EEG networks; Graph theory; PCA; Pattern recognition; Sleep spindle networks.
Copyright © 2018 The Authors. Published by Elsevier B.V. All rights reserved.