Robust Multi-Site ADHD Classification via GraphSAGE-Based Functional Connectivity Modeling from rs-fMRI

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Attention Deficit Hyperactivity Disorder (ADHD) diagnosis relies on behavioral assessment, often leading to delays.
• Resting-state functional magnetic resonance imaging (rs-fMRI) offers potential for objective ADHD diagnosis.
• Existing rs-fMRI studies struggle to fully capture complex brain region interactions.

Purpose of the Study:

• To develop a graph-based deep learning framework for automated ADHD classification using rs-fMRI.
• To combine functional connectivity modeling with graph representation learning for improved diagnostic accuracy.
• To create a scalable and robust model adaptable to multi-site data.

Main Methods:

• Utilized Phase-Locking Value (PLV) for functional connectivity estimation.
• Employed Graph Sample and Aggregate (GraphSAGE) for graph representation learning.
• Integrated regional brain activity and inter-regional interactions for classification.

Main Results:

• The framework demonstrated consistent performance across individual and combined multi-site datasets.
• Achieved high classification metrics: 0.89 Accuracy, 0.96 AUC, and 0.96 Specificity on the combined dataset.
• Outperformed several existing methods in ADHD classification from rs-fMRI data.

Conclusions:

• The proposed framework offers an effective and scalable solution for automated ADHD classification from rs-fMRI.
• Integrating PLV-based connectivity with GraphSAGE learning enhances diagnostic capabilities.
• Contributes to advancing data-driven approaches for neurodevelopmental disorder analysis.