Detecting Community Structure Based on Optimized Modularity by Genetic Algorithm in Resting-State fMRI

Research has attempted to detect the community structure of the brain network using rs-fMRI data to determine differences in brain networks. Traditional clustering methods used to detect the community structure of the brain network, require a priori specification of cluster numbers. However, the cluster number of the brain network remains unknown. In this paper, we propose a new method, GAcut, to detect the community structure of real-world networks and brain functional networks. Here, genetic algorithm is applied to change the connection between nodes, based on optimized modularity Q, and to automatically detect community structure, realizing true, unsupervised analysis. GAcut was then applied to rs-fMRI data to compare differences between autism spectrum disorders (ASDs) and normal controls. Utilizing modularity Q and NMI as measurement indices for differentiation, some characteristic and meaningful network communities that feature in ASDs.