Age Knowledge Guided CNN-Transformer Network for Alzheimer's Disease Diagnosis

Alzheimer's disease (AD) is a neuro-degenerative disorder characterized by subtle structural changes across certain brain regions. These changes can result in a gap between one's chronological age and brain age, which can be served as a biomarker for AD diagnosis. However, it is challenging to leverage this knowledge effectively. To address this problem, we introduced a dual-branch age knowledge guided CNN-Transformer network. The proposed framework comprises three parts: 1) Brain age prediction branch generates predicted brain age and age-related multi-scale feature maps; 2) AD prediction branch fuses the feature maps and extracts AD-related features; 3) Brain Age Guided Attention Module integrates features of the two branches and gives the final AD classification. Furthermore, we propose a novel Channel-Coordinate Attention (CCA) to enhance feature capabilities in 3D neuroimaging data. Our approach leverages the complementary strengths of CNN and Transformer architectures, achieving robust classification results by utilizing age knowledge effectively.