Related Experiment Video
Updated: Sep 28, 2025
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
Using graph convolutional network to characterize individuals with major depressive disorder across multiple imaging
Kun Qin1, Du Lei2, Walter H L Pinaya3
1Huaxi MR Research Center (HMRRC), Department of Radiology, West China Hospital of Sichuan University, Chengdu, China; Department of Psychiatry and Behavioral Neuroscience, University of Cincinnati College of Medicine, Cincinnati, OH, USA; Functional and Molecular Imaging Key Laboratory of Sichuan Province, West China Hospital of Sichuan University, Chengdu, Sichuan, China; Department of Psychiatry and Behavioral Neuroscience, University of Cincinnati College of Medicine, Cincinnati, OH, USA.
Graph convolutional networks (GCNs) show promise for diagnosing major depressive disorder (MDD) by analyzing brain connectivity. This advanced machine learning approach achieved high accuracy in a large, multi-site study, identifying key brain regions involved in MDD.
Area of Science:
- Neuroimaging
- Machine Learning
- Psychiatry
Background:
- Objective biomarkers are crucial for early major depressive disorder (MDD) diagnosis.
- Previous machine learning studies for MDD often used small sample sizes and overlooked brain connectome.
- This study addresses these limitations using a large, multi-site dataset and graph convolutional networks (GCNs).
Purpose of the Study:
- To apply GCNs to a large, multi-site MDD dataset for improved diagnostic accuracy.
- To identify brain regions critical for MDD classification.
- To explore the relationship between brain network topology and clinical features of MDD.
Main Methods:
- Utilized resting-state functional MRI data from 1586 participants (821 MDD, 765 controls) across 16 sites.
- Trained a GCN model on individual whole-brain functional networks to differentiate MDD patients from controls.
- Analyzed salient regions and their topological characteristics in relation to clinical measures.
Main Results:
- GCN achieved 81.5% accuracy (AUC: 0.865), outperforming other classifiers.
- Key regions identified were within the default mode, fronto-parietal, and cingulo-opercular networks.
- Left inferior parietal lobule and left dorsolateral prefrontal cortex topology correlated with depressive severity and illness duration.
Conclusions:
- GCN is a feasible and effective tool for characterizing MDD in large, multi-site studies.
- GCN enhances understanding of MDD neurobiology by detecting functional network disruptions.
- This approach holds potential for improving MDD diagnosis and treatment strategies.
