Comparing Community Detection Algorithms on Neuroimaging Data from Multiple Subjects

It is well-known that the brain is a complex network""brain areas dedicated to different functions. As such,""consisting of""it is natural to shift toward brain network from brain mapping for deeper understanding of brain functions. Although graph theoretical network metrics measuring global or local properties of network topology have been used to investigate the brain network, they do no provide any information about intermediate scale of the brain network, which is provided by the community structure analysis.""In this paper, we propose a method to compare different community detection algorithms for multiple subjects data in terms of the agreement of a group-based community structure with individual community structures. As it is crucial to find a single group-based community structure for a group of subjects to discuss about brain areas and connections, a number of algorithms based on different approaches have been proposed. To show the feasibility of the method for comparing different algorithms, two community detection algorithms based on different approaches ("virtual-typical-subject" and "group analysis") were examined. The Normalized Mutual Information was computed to measure similarity between the group-based community structure and individual community structures derived from resting-state fMRI functional network, and was used for comparing the two algorithms. Our method demonstrated that the algorithm based on the group-analysis approach detected a group-based community structure with greater agreement with individual community structures.