ADHD diagnosis and biomarker detection based on multimodal graph attention convolutional neural network

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Graph neural networks (GNNs) show promise for diagnosing neuropsychiatric disorders using brain functional networks.
• Existing GNN methods for attention-deficit/hyperactivity disorder (ADHD) diagnosis have accuracy limitations due to insufficient multi-modal data integration.

Purpose of the Study:

• To develop an improved GNN model for accurate ADHD identification by effectively integrating multi-modal brain data.
• To enhance the analysis of brain functional networks for ADHD diagnosis.

Main Methods:

• Proposed a graph attention convolutional network (Att-GCN) integrating multi-band amplitude of low-frequency fluctuations (ALFF) and functional connectivity (FC) data.
• Constructed a dual graph representation where ALFF data forms nodes and FC data is converted into supplementary nodes, preserving topological associations.
• Applied graph-based feature selection to refine the dual graph and utilized the Att-GCN for high-level feature learning and ADHD subject identification.

Main Results:

• Achieved an average accuracy of 93.1% in identifying ADHD subjects across ADHD-200 datasets.
• Demonstrated high performance specifically within the right limbic system (11 brain regions).
• Identified reliable ADHD biomarkers located around the thalamus and hippocampus.

Conclusions:

• The proposed Att-GCN model effectively integrates multi-modal brain data for accurate ADHD diagnosis.
• The identified brain regions and biomarkers (thalamus, hippocampus) are crucial for understanding ADHD.
• This approach offers a promising direction for improving diagnostic accuracy in neuropsychiatric disorders using GNNs.