Multimodal Contrastive Enhancement Network for Cross-Ethnic Analysis of Degenerative Brain Regions in Alzheimer’s Disease

Alzheimer’s Disease (AD) exhibits heterogeneous progression patterns across populations with different ethnic backgrounds. However, existing cross-ethnic studies primarily focus on genetic predispositions and demographic risk factors, with limited investigation into neurodegenerative alterations in brain structure and function. Furthermore, conventional approaches often neglect the preprocessing of complementary information across modalities. To address these limitations, we propose a novel framework, named the Multimodal Contrastive Enhancement Network (MCENet), which employs hierarchical attention mechanisms to align inter-modal features and leverages contrastive learning to refine feature representations. By integrating neuroimaging and molecular biomarkers, MCENet identifies discriminative brain regions and improves disease stage prediction. Extensive experiments demonstrate that although both populations exhibit comparable atrophy in regions responsible for memory and emotional regulation, East Asian patients show pronounced degeneration in regions involved in behavioral control, while North American patients are more affected in regions implicated in language and communication. Furthermore, comparative experiments indicate that MCENet consistently outperforms existing methods across four diagnostic metrics for disease stage classification, highlighting its effectiveness and generalizability in neurodegenerative disease diagnosis.