Community detection for directional neural networks inferred from EEG data

One major challenge in neuroscience is to identify the functional modules from multichannel, multiple subjects recordings. Most research on community detection has focused on finding the association matrix based on functional connectivity, instead of effective connectivity, thus not capturing the causality in the network. In this paper, we propose a community detection algorithm suitable for weighted and asymmetric (directed) networks representing effective connectivity, and apply the algorithm to multichannel electroencephalogram (EEG) data. In addition, we extend the algorithm to find one common community structure from multiple subjects.

I. Introduction

With the advance of neuroimaging technology, such as EEG, it is possible to record brain activity with higher temporal resolution and accuracy than ever before. In order to understand the functioning of the brain better, methods to identify the communities or functional modules from the observed multichannel, multiple subjects recordings have been developed. Functional and effective connectivity are two widely studied measures to quantify the connectivity patterns in the brain. Unlike functional connectivity which only quantifies the statistical dependencies between two processes, effective connectivity quantifies the influence one node exerts on another node. Traditionally, effective connectivity has been quantified using measures of causality, such as Granger causality and partial directed coherence (PDC) [1]. Granger causality based methods are model dependent and limited to detecting the linear relations, however, EEG recordings are known to have nonlinear dependencies between recordings from different sites. Therefore, in our previous work, directed information was proposed to quantify the information flow in the brain network [2]. Unlike Granger causality, directed information is model free and can quantify the nonlinear relations.