An FDR-controlled, exploratory group modeling for assessing brain connectivity

Recent development in the inference of brain connectivity from neuroimaging data such as functional magnetic resonance images (fMRI) provides better understanding of brain activities and functions. The group analysis of fMRI data usually focuses on functional connectivity, while exploratory graphical modeling of effective connectivity is generally designed for the single-subject case. In this paper, a new group-level network learning method, the group PC_fdr algorithm, is proposed by combining group inference with exploratory graphical modeling for estimating brain connectivity, with inter-subject variances taken into consideration. The simulation results demonstrate that this group method can accurately recover underlying group-level structures and steadily control the false discovery rate at the user specified level. It is also applied to a real fMRI dataset of normal subjects and Parkinson´s Disease patients to infer the brain connectivity at the group level.