Graph Filtering for Hub Node Identification in Brain Networks

Over the past two decades, complex network theory has been used to model both the functional and structural connectivity of the brain. Different graph theoretic metrics have been used to characterize the topology of brain networks. One important characteristic of brain networks is the existence of hub nodes, which are central to neural integration. Previous methods for hub node identification are primarily based on computing the centrality of the nodes based on the functional connectivity network. However, these approaches to hub identification may identify portions of large brain systems rather than critical nodes of brain networks. In this paper, we introduce an alternative view of hub nodes utilizing both the functional connectivity network and the neurophysiological signals defined on the nodes of this network. An unsupervised learning method is proposed to learn a graph filter that separates the hub nodes from normal nodes, where the hub nodes are defined based on the homophily of a node with respect to its neighbors. A metric to quantify the local total variation is introduced to identify the possible hub nodes. The proposed method is applied to electroencephalogram (EEG) data collected from a study of error monitoring in the human brain. The detected hub nodes are compared to existing methods.