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

The EEG signal and its oscillatory components may relate with temporal modulation of information processing of a sensory activation in a local electrical field 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 was designed. The functional coupling among EEG electrodes were quantified via mutual information in the time-frequency plane. A spectral clustering algorithm was used to parcellate memory related circuits in the brain in a load-dependent manner. The method is 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 offers to construct significant clusters without fixing a specified number of clusters and increases the inside cluster significance by normalized-cut value decomposition at each clustering level. In the N-cut clustering results, clustered nodes which are projected on occipital and bilateral regions increase with respect to the memory load. In the soft clustering results, inter-cluster connections between left lateral and occipital clusters are decreasing in the second (1250ms) time interval which can be linked to the enhancement of posterior region due to the increase in the memory demand.