A sparse transformer generation network for brain imaging genetic association

Abstract

Brain imaging and genetic data are commonly utilized to investigate brain aging and diseases, particularly Alzheimer’s Disease (AD). Imaging genetics analyze the associations between neuroimaging and genetic data to reveal potential pathological mechanisms for more accurate diagnosis. Many existing methods have limitations in performing fine-grained association analysis between genetic data and phenotypic features extracted from predefined regions of interest in imaging data. To address this issue, this paper proposes a sparse transformer association analysis (STAA) framework that integrates phenotype and genotype feature extraction, identification, and association analysis into a unified model. A key component of the framework is a cross-modal generation network that connects genetic variants with imaging data, enhancing the understanding of the genetic associations underlying imaging patterns in AD and brain aging. Validated using the simulated data and real ADNI dataset, STAA shows superior performance in age regression and AD diagnosis, identifying key genetic and imaging biomarkers and providing association analysis of genetic imaging expression patterns at the voxel level.