A Novel Sparse Overlapping Modularized Gaussian Graphical Model for Functional Connectivity Estimation

摘要

Neural mechanisms underlying brain functional systems remain poorly understood, the problem of estimating statistically robust and biologically meaningful functional connectivity by limited functional magnetic resonance imaging (fMRl) time series containing complex noises remains an open field. Addressing this issue, motivated by recent studies, which have highlighted that brain existing functional overlapping modularized patterns, we propose a novel sparse overlapping modularized Gaussian graphical model (SOMGGM) that estimates functional connectivity by modularizing the connection patterns and allowing each brain region belonging to multiple modules. Extensive experimental results demonstrate that the proposed SOMGGM not only has more power to accurately estimate functional connectivity network structure, but also improves feature extraction and enhances the performance in the brain neurological disease diagnosis task. Additionally, SOMGGM can help to find the brain regions assigned to multiple network modules which are likely important hub nodes. In general, the proposed SOMGGM offers a new computational methodology for brain functional connectivity estimation.