Adaptive Thresholding of Functional Connectivity Networks for fMRI-Based Brain Disease Analysis

Abstract

Functional connectivity (FC) networks based on functional magnetic resonance imaging (fMRI) data have been widely applied to automated identification of brain diseases, such as attention deficit hyperactivity disorder (ADHD) and Alzheimer’s disease (AD). To generate compact representations of FC networks for disease analysis, various thresholding strategies have been developed for analyzing brain FC networks. However, existing studies typically employ predefined values or percentages of connections to threshold the whole FC networks, thus ignoring the diversity of temporal correlations (particularly strong correlations) among different brain regions. In addition, in practice, it is usually very challenging to decide the optimal threshold or connection percentage in FC network analysis. To address these problems, in this paper, we propose a weight distribution based thresholding (WDT) method for FC network analysis with resting-state function MRI data. Specifically, for FC between a pair of brain regions, we calculate its optimal threshold value by using the weight (i.e., temporal correlation) distributions of the FC across two subject groups (i.e., patient and normal groups). The proposed WDT method can adaptively yields FC-specific thresholds, thus preserving the diversity information of FCs among different brain regions. Experiment results on both ADNI and ADHD-200 datasets with rs-fMRI data demonstrate the effectiveness of our proposed WDT method.