Deep Generative Modelling of Human Brain Connectome

This thesis explores brain disorder classification using AI techniques based on human brain connectomes, encompassing both structural (SC) and functional (FC) connectivity. Although AI holds great promise in brain research, its reliance on large, costly datasets poses significant challenges. To overcome this, we developed generative models to produce realistic brain connectivity data. These models create both static and dynamic FC, simulate white matter streamlines from MRI scans, and translate between SC and FC when one modality is unavailable. By integrating these methods for connectome generation, translation, and classification, we achieved substantial improvements in brain disorder diagnosis.