Research on Intelligent Diagnosis Integrating Disease Subgraph and Large Language Model

AI-driven diagnostic tools proved to be an effective aid for healthcare professionals in predicting diseases, thereby streamlining healthcare services. In order to improve the accuracy and efficiency in the field of intelligent diagnosis, we used Chinese electronic medical records as the data base, incorporated knowledge graphs as the external knowledge source, and introduced a large language model to reason about electronic medical records. To solve the problems of difficulty in knowledge graph construction and insufficient semantic relationship extraction, we proposed an intelligent diagnosis method that combined disease subgraphs with a large language model (DeLL). The DeLL model constructed prompt templates using the text of the electronic medical record, guided the large language model to generate new descriptions, and combined these with the multi-granularity information in the EMR to obtain enhanced textual representations. Meanwhile, the DeLL model constructed disease subgraphs by utilizing the target diseases and the knowledge graph, and encoded them using graph convolutional neural networks to fuse the disease knowledge representations with the LLM-enhanced text representations for disease prediction. The model was experimentally evaluated on a dataset consisting of multi-label and single-label electronic medical records and achieved F1 micro scores of 85.01% and 93.19%, respectively, representing improvements of 3.20% and 1.80% over the optimal baseline results.