A dynamic graph convolutional neural network framework reveals new insights into connectome dysfunctions in ADHD

Area of Science:

• Neuroscience
• Computer Science
• Medical Imaging

Background:

• The pathological mechanisms of attention deficit hyperactivity disorder (ADHD) remain unclear, hindering precise diagnosis.
• Functional magnetic resonance imaging (fMRI) is crucial for studying brain functional connectomes.
• Existing methods often fail to fully utilize the topological information within brain networks.

Purpose of the Study:

• To develop a novel deep learning approach for improved ADHD diagnosis.
• To leverage topological information from brain functional connectomes.
• To enhance the precision of ADHD diagnosis using neuroimaging data.

Main Methods:

• Proposed a dynamic graph convolutional network (dGCN) model.
• Trained the dGCN with sparse brain regional connections and dynamic graph features.
• Developed a novel convolutional readout layer for improved graph representation.
• Utilized fMRI data for brain connectome analysis.

Main Results:

• The dGCN model significantly outperformed existing machine learning and deep learning methods in ADHD diagnosis.
• Identified functional abnormalities in specific brain regions including the temporal pole, gyrus rectus, and cerebellar gyri.
• Observed a positive correlation between identified connectomic abnormalities and ADHD symptom severity.

Conclusions:

• The proposed dGCN model shows significant promise for functional network-based precision diagnosis of ADHD.
• The approach is broadly applicable to the study of other mental disorders using brain connectomes.
• Highlights the importance of topological information in characterizing brain disorders.