Mediation model has been employed in brain disorders to detect the underlying mechanisms between genetic variants and diagnostic outcomes implicitly mediated by intermediate imaging biomarkers. However, the statistical power is influenced by the modest effect of individual genetic variant on both diagnostic and imaging phenotypes and the limited sample size with all genotyping, diagnosis and imaging data available. In this study, we propose the polygenic mediation analysis that comprises polygenic risk score (PRS) to aggregate genetic effects of a set of candidate variants and then explore the implicit effect of imaging phenotypes between PRS and disease status. We applied our proposed method to an amyloid imaging genetic study of Alzheimer's disease, identified multiple imaging mediators linking PRS with AD, and further demonstrated the out-performance of PRS on mediator detection than individual variant alone.