Multimodal Representations Learning and Adversarial Hypergraph Fusion for Early Alzheimer’s Disease Prediction

Multimodal neuroimage can provide complementary information about the dementia, but small size of complete multimodal data limits ability in representation learning. Moreover, distribution inconsistency from different modalities may lead to ineffective fusion, which fails sufficiently explore intra-modal and inter-modal interactions compromises disease diagnosis performance. To solve these problems, we proposed a novel learning adversarial hypergraph fusion (MRL-AHF) framework for Alzheimer’s using trimodal images. First, strategy pre-trained model are incorporated into MRL extract latent representations data. Then two hypergraphs constructed network based on graph convolution is employed narrow difference hyperedge features. Finally, hyperedge-invariant features fused prediction by convolution. Experiments public Disease Neuroimaging Initiative(ADNI) database demonstrate that our achieves superior performance detection compared with other related models provides possible way understand underlying mechanisms disorder’s progression analyzing abnormal brain connections.