The efficacy of topological properties of functional brain networks in identifying major depressive disorder

Abstract

Major Depressive Disorder (MDD) is a common mental disorder characterized by cognitive impairment, and its pathophysiology remains to be explored. In this study, we aimed to explore the efficacy of brain network topological properties (TPs) in identifying MDD patients, revealing variational brain regions with efficient TPs. Functional connectivity (FC) networks were constructed from resting-state functional magnetic resonance imaging (rs-fMRI). Small-worldness did not exhibit significant variations in MDD patients. Subsequently, two-sample t-tests were employed to screen FC and reconstruct the network. The discriminative ability of TPs between MDD patients and healthy controls was analyzed using receiver operating characteristic (ROC), ROC analysis showed the small-worldness of binary reconstructed FC network (p < 0.05) was reduced in MDD patients, with area under the curve (AUC) of local efficiency (Le) and clustering coefficient (Cp) as sample features having AUC of 0.6351 and 0.6347 respectively being optimal. The AUC of Le and Cp for retained brain regions by T-test (p < 0.05) were 0.6795 and 0.6956 respectively. Further, support vector machine (SVM) model assessed the effectiveness of TPs in identifying MDD patients, and it identified the Le and Cp in brain regions selected by the least absolute shrinkage and selection operator (LASSO), with average accuracy from leave-one-site-out cross-validation being 62.03% and 61.44%. Additionally, shapley additive explanations (SHAP) was employed to elucidate variations in TPs across brain regions, revealing that predominant variations among MDD patients occurred within the default mode network. These results reveal efficient TPs that can provide empirical evidence for utilizing nodal TPs as effective inputs for deep learning on graph structures, contributing to understanding the pathological mechanisms of MDD.