Functional parcellation of memory related brain networks by spectral clustering of EEG data

Abstract

The EEG signal and its oscillatory components may relate with temporal modulation of information processing of a sensory activation in a local electrical eld and neural populations. In this study, we investigate the clustering information of alpha band brain networks during memory load task. For this purpose, short time memory experiment with a varying memory load combinations was designed. The functional coupling among EEG electrodes were quanti ed via mutual information in the timefrequency plane. A spectral clustering algorithm was used to parcellate memory related circuits in the brain in a load-dependent manner. The method was based on determining the eigenspectrum of the adjacency matrix of a graph and assigning nodes to clusters with respect to this spectrum. To be able to circumvent the problem of choosing the number of clusters beforehand a soft clustering approach was implemented. It is a novel method which allows to construct signi cant clusters without xing their number and increases the inside cluster signi cance by normalized-cut value decomposition at each clustering level. In the N-cut clustering, clustered nodes which are projected on occipital and bilateral regions increase in number with respect to the memory load. In soft clustering, inter-cluster connections between left lateral and occipital clusters are decreasing in the second time interval which can be linked to the enhancement of posterior region due to an increase in the memory demand.|Keywords : Normalized Cut, EEG, Working Memory, Memory Load, Mutual Information, Information Theory.