Identifying Biomarkers of Subjective Cognitive Decline Using Graph Convolutional Neural Network for fMRI Analysis

Abstract

Subjective cognitive decline (SCD) is the preclinical stage of Alzheimer’s disease (AD). People with SCD have a higher chance of developing mild cognitive impairment and AD than those aging normally. In the present study, we collected resting state functional magnetic resonance imaging (rs-fMRI) data for 69 patients with SCD and 75 normal controls (NC); using statistical analysis, a support vector machine (SVM), and graph convolutional neural networks (GCNs), we examined the brain-related differences between patients with SCD and NC. Clinical scale scores show the best distinguishing ability between patients with SCD and NC, and we further used the two-sample t-test, SVM, and GCN model with an attention mechanism to obtain the top 10 brain regions contributing to performance on recognition tasks. The results showed that the thalamus, and cingulum in the Anatomical Automatic Labeling template showed significant differences between patients with SCD and NC. We further discussed the roles of these identified brain regions in the diagnosis of SCD and AD. Our research thus provided statistical evidence that can aid in identifying early-stage AD.