Adaptive Multi-Scale Dynamic Graph Representation Learning With Overlapping Community-Awareness for ASD...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Dynamic functional connectivity (dFC) is crucial for brain disease diagnosis.
• Graph neural networks (GNNs) leverage brain topology for dFC analysis.
• Existing methods face limitations in capturing multi-scale temporal dynamics and multi-network region participation.

Purpose of the Study:

• To propose Ada-MST, an adaptive multi-scale spatio-temporal model for brain disease diagnosis.
• To address limitations in conventional sliding window approaches and GNN representations.
• To enhance diagnostic accuracy by incorporating subject-specific temporal characteristics and multi-network region participation.

Main Methods:

• Developed an adaptive multi-scale spatio-temporal model (Ada-MST).
• Constructed personalized multi-scale dFC graphs adapting to subject-specific temporal dynamics.
• Introduced an overlapping community-aware readout module for improved graph-level representations, considering multi-network region participation.

Main Results:

• Ada-MST demonstrated superior performance compared to state-of-the-art methods on ABIDE-I and ABIDE-II datasets.
• Visualization confirmed the generalizability of subject-adaptive graphs and their focus on disease-related brain activity.
• Fuzzy community memberships revealed distinct patterns across diseases, highlighting potential biomarkers.

Conclusions:

• Ada-MST offers an advanced approach for brain disease diagnosis using dFC.
• The model's ability to capture multi-scale spatio-temporal features and multi-network participation enhances diagnostic accuracy.
• Functional community membership analysis shows promise for identifying disease-specific biomarkers.