Trusted Multi‐View Graph Convolutional Network With Cross‐View GraphMAE for Alzheimer's Disease Classification

Low‐order and high‐order functional connectivity (LOFC and HOFC) provide complementary information for Alzheimer's disease (AD) classification. However, existing multi‐view methods usually assign fixed view weights, which makes their performance suboptimal when view quality varies across subjects. To address these challenges, we propose a novel Trusted Multi‐view Graph Convolutional Network with Cross‐view Graph Masked Autoencoder (CGM‐TMVGCN) for AD classification. The proposed framework includes three main innovations. First, an evidence‐level fusion is proposed for trusted multi‐view GCN, enabling adaptive integration of LOFC and HOFC views according to data quality. Second, to alleviate the limited‐sample problem, we develop a Cross‐view Graph Masked Autoencoder, which pretrains LOFC and HOFC graph encoders by reconstructing masked nodes across views. Third, a prototype‐guided contrastive learning strategy is introduced to improve the discrimination of LOFC and HOFC evidence vectors. Experiments on the ADNI database show that CGM‐TMVGCN achieves an F1‐macro of 63.45 ± 2.60, an F1‐weighted of 74.28 ± 1.86, and an AUC of 88.75 ± 0.62, outperforming competing multi‐view GCN and graph self‐supervised learning methods. These results indicate that CGM‐TMVGCN effectively integrates LOFC and HOFC features for reliable AD classification.