A typical sample-driven learning framework for automatic disease diagnosis

Chenwei Yan, Xinxin You, Xiangling Fu, Xien Liu, Ji Wu

Abstract: Highlights•Proposes TD-GNN, a framework to learn knowledge representation from samples.•Alleviates the imbalanced classification problem of diagnoses and diagnostic sub-types.•Enhances distinguishable feature learning with self-supervised contrastive strategy.•Contructs a real-world disease diagnosis dataset covering 350 diseases and 122,464 EHRs.•Extensive experiments demonstrate the framework’s effectiveness and interpretability.