Mediation modeling has become an important tool in the psychological research for studying the intermediate effect of neuroimaging biomarkers on the causal pathway from genetic variations to diagnostic outcomes. One the other side, researchers begin combining of multiple types of measurements, which are scientifically complementary, to strengthen our understanding of brain dynamics and their associations with neurological disorders. However, little work has been done when the mediator is high-dimensional multimodal neuroimaging data. We propose a mediation model framework with functional connectivity data and multiple region-level imaging data as mediators, and we conduct model estimation by imposing group sparsity penalty and graph based Laplacian penalty. We illustrate our method with a multimodal brain pathway analysis having both positron emission tomography (PET) and magnetic resonance imaging (MRI) measurements as mediators in the association between Alzheimer's Disease Assessment Scale–Cognitive Subscale (ADAS-Cog) and single nucleotide polymorphisms (SNPs) in the APOE gene, identifying which brain locations in each modality mediate the relationship.