Alzheimer's disease (AD) is a highly heritable neurodegenerative disease, and brain imaging genetics (BIG) has become a key area for understanding its pathogenesis. However, existing methods often ignore the complex interrelationships between the multiple factors that lead to AD, especially when exploring the intrinsic connection between brain imaging features and gene variation. To address this challenge, we proposed a multimodal association modeling framework (MAM-GAN) based on generative adversarial networks, which aims to deeply reveal the association between genes and brain imaging features and apply it to disease state prediction. To verify the effectiveness of the framework, we conducted experiments using public datasets, and the results showed that MAM-GAN performed well in two classification tasks and successfully identified biomarkers closely related to AD.