ADHD Classification with GCN via Joint Feature Learning among Nodes and Edges

Comprehensive Summary

Wang et al. studied the improvements of modeling node-edge features in brain functional connectivity networks for the classification and biomarker identification of attention-deficit/hyperactivity disorder (ADHD). The researchers proposed a graph convolutional network, called JNEL-GCN, that alternates between capturing regional brain activity with multi-band ALFF features and functional connectivity and inter-node relationships from rs-fMRI. The model was tested against the ADHD-200 dataset, yielding 97.3% accuracy, and the ABIDE-I dataset, yielding 97.1% accuracy, thus outperforming other deep learning, GCN-based, and binary hypothesis methods. Gradient analysis found abnormal areas in the default mode network and the bilateral limbic networks. These findings are similar to the results found in the past about ADHD. The researchers have highlighted that the improved accuracy in classification can be obtained by modifying the node and edge interaction.

Outcomes and Implications

This study is relevant because the current process used in the diagnosis of ADHD is largely behavioral, resulting in the misdiagnosis of the disorder. By using rs-fMRI data and modeling both regional activity and inter-regional connectivity, JNEL-GCN provides an objective way for identifying brain network deviations that are related to ADHD. This approach could improve earlier clinical diagnosis, diagnostic confidence, and aid in identifying different subtypes of ADHD, allowing for a more personalized treatment. While the model has not yet been tested in clinical settings, the researchers suggest that continued testing and application of neuroimaging data could make it a significant supporting tool to make decisions in neuropsychiatric research and clinical practice.