Brain multigraph prediction using topology-aware adversarial graph neural network

Brain graphs (i.e, connectomes) constructed from medical scans such as magnetic resonance imaging (MRI) have become increasingly important tools to characterize the abnormal changes in human brain. Due high acquisition cost and processing time of multimodal MRI, existing deep learning frameworks based on Generative Adversarial Network (GAN) focused predicting missing images a few modalities. While brain help better understand how particular disorder can change connectional facets brain, synthesizing target multigraph multiple graphs) single source graph is strikingly lacking. Additionally, generation works mainly learn one model for each domain which limits their scalability jointly domains. Besides, while they consider global topological scale (i.e., connectivity structure), overlook local topology at node (e.g., central graph). To address these limitations, we introduce topology-aware GAN architecture (topoGAN), predicts preserving structure graph. Its three key innovations are: (i) designing novel adversarial auto-encoder one, (ii) clustering encoded order handle mode collapse issue proposing cluster-specific decoder, (iii) introducing loss force prediction topologically sound graphs. The experimental results using five domains demonstrated outperformance our method comparison with baseline approaches.