Generative AI Empowers Estimation of Brain Structure-Function Connectivity for Alzheimer's Disease

Fusing structural and functional brain image has become a hot topic in estimating multimodal effective connectivity for cognitive disorder identification. However, it is challenging to effectively and accurately integrate the complementary causal connectivity information using structural and functional brain images. In this work, a novel incidence-aware transformer generative adversarial network (IT-GAN) is proposed to estimate effective connectivity from resting-state functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI). IT-GAN unifies cross-modal alignment, progressive incidence-aware fusion, and multi-aspect consistency enforcement to generate robust, directionally discriminative multimodal effective connectivity. Experimental results demonstrate the proposed model's better performance in AD identification compared with other related works. The IT-GAN can accurately estimate effective connectivity from fMRI and DTI, and has the potential to revolutionize understanding the mechanism of the cognitive disease, which leads to new insights into AD's pathological study and early treatment.