Objective: To address the challenges posed by the rapid growth of medical data and the fragmentation of knowledge, this study aims to construct a medical knowledge graph (KG) and provide efficient knowledge services for clinical practice. Methods: A total of 44,157 entities and 291,170 relationships from an open-source database were integrated to build a local medical KG based on Neo4j. Graph algorithms including degree centrality, Louvain community detection, K-nearest neighbor, and Dijkstra’s algorithm were applied to analyze the data. The retrieval results from the KG were combined with the Spark Lite model from IFlytek to develop a dual-channel question-and-answer system. Results: Highly related entities, such as acute urethritis and blood routine tests, were successfully identified. The analysis yielded 35 disease communities and 17 department communities. Highly similar disease pairs, such as “lung abscess” and “pulmonary bullae,” were discovered. Potential therapeutic pathways, such as “Erythromycin Ethylsuccinate Granules - Erythrasma,” were uncovered, revealing clinical associations among various entities. The system is accessible at http://yangbiolab.cn:8054/ . Conclusion: Graph algorithms effectively mine key patterns and potential associations within medical knowledge, with several findings aligning closely with clinical practice. The integrated system offers an intuitive platform for exploring medical knowledge.
Huang et al. (Tue,) studied this question.