Multi-Modal Diagnosis of Alzheimer's Disease Using Interpretable Graph Convolutional Networks.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Neurological diseases like Alzheimer's disease (AD) are characterized by altered brain connectivity.
• Graph convolutional networks (GCNs) show promise in analyzing brain networks for disease detection.
• Integrating multiple imaging modalities can enhance the accuracy of GCN-based disease identification.

Purpose of the Study:

• To develop and evaluate a multi-modal sparse interpretable GCN framework (SGCN) for detecting Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• To identify disease-specific brain regions of interest (ROIs) and network connections using SGCN.
• To assess the potential of SGCN for developing novel biomarkers and enabling precision diagnostics for AD/MCI.

Main Methods:

• Proposed a multi-modal sparse interpretable GCN framework (SGCN).
• Utilized sparse regional importance probability to identify signature ROIs.
• Employed connective importance probability to reveal disease-specific brain network connections.
• Evaluated SGCN on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database using multi-modal brain images.

Main Results:

• SGCN effectively learned ROI features that enhanced AD status identification.
• Identified brain abnormalities were significantly correlated with AD-related clinical symptoms.
• Interpreted brain dysfunctions at the level of large-scale neural systems and sex-related connectivity abnormalities in AD/MCI.

Conclusions:

• The developed SGCN framework is effective for multi-modal diagnosis of AD and MCI.
• Identified salient ROIs and prominent brain connectivity abnormalities are crucial for novel biomarker development.
• Findings contribute to understanding network-based disorders and offer potential for precision diagnostics.