Local-Global Co-Attention Neural Network for Alzheimer’s Disease Diagnosis

Deep neural network (DNN) models have achieved great success in automatic Alzheimer’s Disease (AD) diagnosis with 3D brain medical images. However, most existing methods seldom consider the correlations between brain image patterns and non-imaging clinical information that is usually adopted by medical professionals in practice. To exploit the heterogeneous features from both imaging and non-imaging data, we propose a Local-Global Co-attention neural network (LGC-Net), which features two specialized blocks: the Paired Local-Global Attention (PLGA) Block and the Paired Local-Global Cross Attention (PLGCroA) Block, focusing on capturing local and global features from two modalities. Furthermore, LGC-Net employs a joint learning approach, combining supervised learning for category-level discriminative abilities with contrastive learning to maintain task-relevant information. Our experiments on the ADNI dataset show that LGC-Net surpasses existing methods, demonstrating its effectiveness in fusing visual and tabular modalities for AD diagnosis.