A dual path graph neural network framework for dementia diagnosis

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Dementia is characterized by neural pathway damage and neuronal connection degeneration.
• Graph neural networks (GNNs) are used for brain network modeling.
• Integrating diverse features (temporal, spatial, spectral) for diagnosing neurocognitive disorders remains a challenge.

Purpose of the Study:

• To develop an advanced model for diagnosing neurocognitive disorders by effectively integrating multi-modal brain network features.
• To address the limitations of existing methods in leveraging complementary temporal, spatial, and spectral information.

Main Methods:

• Proposed a novel Bi-path Multi-scale Graph Neural Network (Bi-MCGNN).
• Integrated temporal-spatial and spatial-frequency pathways within a unified framework.
• Utilized specialized correlation matrices for enhanced graph representation.
• Applied multi-scale graph convolution and an attention mechanism for comprehensive feature analysis.

Main Results:

• The Bi-MCGNN model successfully integrated diverse brain features.
• Multi-scale analysis captured connectivity patterns at various resolutions.
• The attention mechanism effectively enhanced cross-domain features.
• Achieved superior performance compared to state-of-the-art methods on two real-world datasets.

Conclusions:

• The proposed Bi-MCGNN offers a powerful and effective approach for diagnosing neurocognitive disorders.
• Integrating multi-modal brain network features significantly improves diagnostic accuracy.
• The model demonstrates the potential of advanced GNN architectures in neuroimaging analysis.