Research on Double-Graphs Knowledge-Enhanced Intelligent Diagnosis

Intelligent diagnosis is an effective method to assist doctors in disease diagnosis. Integrating domain knowledge graphs into the intelligent diagnosis process can enhance diagnostic effect. The enhancement effects of different knowledge in the knowledge graph are different. The knowledge introduced based on electronic medical record text can explain medical terms, extract hidden relations, and enrich text representation. However, this method is prone to generating noise and affecting the diagnostic results. Disease is our diagnostic goal, introducing external knowledge centered around diseases can not only obtain richer and more professional disease-related information, but also make the introduced knowledge more accurate and reduce the impact of external knowledge noise. Therefore, this paper proposes a Double-Graphs Knowledge-Enhanced Intelligent Diagnosis Model (DGKE). Firstly, we extract the knowledge related to electronic medical record text from the knowledge graph and construct a text subgraph. At the same time, we obtain the knowledge associated with the disease to be diagnosed and construct a disease subgraph. Then, the two graph representations are fused using a light-attention to obtain an external knowledge representation for the disease to be diagnosed. Finally, the disease-oriented knowledge representation is fused with the hierarchical information-enhanced text representation to obtain the knowledge-enhanced text representation, which is mapped to the disease list space to be diagnosed for prediction. Experiments are conducted on the COEMRs (Chinese Obstetric Electronic Medical Records) and the C-EMRs (Chinese Electronic Medical Records). Compared with models without disease knowledge enhancement, the F1_micro increase by 0.65% and 1.44% respectively and the F1_macro increase by 4.06%, 2.23% respectively.