Stage-aware Brain Graph Learning for Alzheimer's Disease

Current machine learning-based Alzheimers disease (AD) diagnosis methods fail to explore the distinctive brain patterns across different AD stages, lacking the ability to trace the trajectory of AD progression. This limitation can lead to an oversight of the pathological mechanisms of AD and suboptimal performance in AD diagnosis. To overcome this challenge, this paper proposes a novel stage-aware brain graph learning model. Particularly, we analyze the different brain patterns of each AD stage in terms of stage-specific brain graphs. We design a Stage Feature-enhanced Graph Contrastive Learning method, named SF-GCL, utilizing specific features within each AD stage to perform graph augmentation, thereby effectively capturing differences between stages. Significantly, this study unveils the specific brain patterns corresponding to each AD stage, showing great potential in tracing the trajectory of brain degeneration. Experimental results on a real-world dataset demonstrate the superiority of our model.