A Hybrid Point-Sets Clustering Approach to Identification of Resting State Functional Network

Study of functional brain networks via resting state fMRI (rsfMRI) data has received increasing interest in the literature recently. Data-driven voxel-wise clustering approaches have been the mainstream approaches to automatic identification of resting state functional networks (rsFN). A basic assumption in these methods is that within each rsFN, voxels should be highly correlated with each other in terms of rsfMRI time courses. In this paper, we relax this assumption and assume that voxels in each rsFN do not necessarily have similar rsfMRI time series, but have either spatial vicinity or functional coherence. Consequently, we designed and implemented a hybrid point-sets clustering approach to identifying rsFNs based on rsfMRI data. Our experimental results show that this hybrid point-sets clustering method can identify meaningful and consistent rsFNs such as default mode network and motor network across subjects.