Spectral Graph Neural Network-Based Multi-Atlas Brain Network Fusion for Major Depressive Disorder Diagnosis

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Major Depressive Disorder (MDD) significantly impacts global health.
• Functional Magnetic Resonance Imaging (fMRI) shows potential for objective MDD diagnosis.
• Current fMRI methods often use single brain atlases, limiting the capture of multi-scale brain network complexity.

Purpose of the Study:

• To develop a novel multi-atlas fusion method for enhanced MDD diagnosis.
• To address the limitations of single-atlas approaches in capturing complex functional brain networks.
• To improve the accuracy and reliability of MDD detection using fMRI data.

Main Methods:

• Proposed a unified framework integrating early and late fusion techniques.
• Introduced the holistic Functional Connectivity Network (FCN) to capture intra- and inter-atlas relationships across different brain parcellation scales.
• Utilized multiple spectral Graph Convolutional Neural Networks (GCNNs) and decision-level ensembles for decoding FCN and improving diagnostic performance.

Main Results:

• The novel multi-atlas fusion method demonstrated superior performance compared to single- and existing multi-atlas methods.
• The holistic FCN effectively identified potential disease-related patterns associated with MDD.
• The approach showed robust performance across different baseline models on public resting-state fMRI datasets.

Conclusions:

• The proposed multi-atlas fusion framework offers a more reliable technique for MDD diagnosis.
• This method enhances the accuracy of MDD detection by capturing multi-scale functional brain connectivity.
• The findings suggest a significant advancement in leveraging fMRI for early and accurate diagnosis of Major Depressive Disorder.