A Connectivity-Enhanced Concatenated Dynamic Graph Neural Network for Cognitive Ability Prediction

Graph neural networks have gained extensive application in the analysis of brain function and recognition of neural biomarkers due to their proficiency in processing graphstructured data. Here, we propose a Connectivity-Enhanced Concatenated Dynamic Graph Isomerism Network (C²DGIN) model. This model predicts cognitive abilities by constructing dynamic graph representations with spatiotemporal characteristics and connectivity, and by concatenating multiple features in graph representation learning. We validate the C²DGIN model on resting-state functional magnetic resonance imaging (rs-fMRI) data sourced from the Human Connectome Project (HCP) dataset. The results indicate the superiority of C²DGIN in predicting cognitive abilities. In addition, our study also demonstrated a significant correlation between cognitive task performance and functional connectivity in resting state. This model can objectively reflect individual cognitive ability and thus provides support for future reliable cognitive assessment test.