Multi-Region Correlation Based Functional Brain Network for Disease Diagnosis and Cognitive States Detection

A functional brain network has attracted much attention due to its capability of characterizing the functional connectivity patterns of the brain. The existing methods for network construction usually rely on the conventional measures such as Pearson correlation to determine the pairwise similarity between each pair of brain regions, thus ignoring the global structure relationship and the information communication flow among different brain regions. To this end, this paper proposes a directional brain network construction method based on effective distance. Specifically, the effective distance can capture the hidden relevance pattern among all the brain regions and provide a directional functional network reflecting information propagation paths among the functional brain areas simultaneously. When estimating the probability of the direction and the strength of the connectivity between two regions, the structure information characterized by their neighbors is also considered in our method. The functional brain network produced by our method is more flexible in uncovering physiological mechanisms of the brain compared with the conventional undirected network. The experiments on two fMRI data analysis tasks, i.e., disease diagnosis and cognitive state detection, show that our method outperforms the conventional functional brain network approaches, including Pearson correlation, structural equation modeling, and sparse representation-based method.