Motif-Matching Based Sub-Braingraph Level Networks for Noisy Resting-State fMRI Analysis

Abstract

Biomarkers extracted from rs-fMRI based brain functional connectivity (FC) can assist in diagnosing various brain disorders. Recently, several graph-based methods have been proposed for modeling the braingraph of brain disorders and brain disorders diagnosis. However, those methods overlook the subgraph structure of brain-graph and the noise in rs-fMRI. In light of current deficiencies, this paper proposed a Motif-Matching based Sub-Braingraph Level network (MMS-Net) with Residual Shrinkage Denoising module for rs-fMRI modeling and brain disorder diagnosis. Comprehensive experiments performed on multi-site databases show that the accuracy of MMS-Net surpasses many state-of-art methods, which proves our framework can extract braingraph features more precisely and has potential for auxiliary diagnosis of brain disorders in clinical settings. We further visualize and analyze the brain network of several brain disorders on multi-levels and discover several meaningful and interpretative biomarkers, which can facilitate the development of biomarkers for auxiliary diagnosis of brain disorders.