This study investigates the neural mechanisms underlying exercise addiction by analyzing resting-state brain functional connectivity. The study develops a multimodal brain network identification method that combines K-means clustering with Graph Convolutional Networks (GCNs). The approach integrates white-matter connectivity, gray-matter structural features, and functional activity, forming a graph-based representation framework capable of cross-modal feature fusion. This framework provides a technical foundation for applying generative artificial intelligence to brain-imaging pattern modeling and establishes a methodological basis for developing generative models of brain functional features associated with exercise addiction. Participants are recruited from Guangzhou Sport University. Based on the Exercise Addiction Inventory (EAI), six individuals (13.04%) are identified as exhibiting exercise-addiction symptoms or risk and undergo magnetic resonance imaging. The analysis reveals a significant negative correlation between gray-matter volume and functional activity in the left supplementary motor area (SMA) (r = -0.22, p < 0.05). Furthermore, the gray-matter volume of the left SMA mediates the relationship between its functional activity and exercise addiction (95% CI = 0.044, 1.114, p < 0.05). The small sample size of the exercise-addiction group limits statistical power and generalizability, and functional connectivity differences require validation in larger cohorts. Nevertheless, the findings reveal key brain-region characteristics associated with exercise addiction and provide preliminary evidence for applying multimodal generative artificial intelligence to study brain functional mechanisms. Overall, this study establishes a systematic, graph-based framework for exploring complex brain networks and offers a foundation for future research on individualized modeling and cross-modal pattern reconstruction in exercise-addiction neuroscience.
Xie et al. (Tue,) studied this question.