Research Of Automated Disease Diagnosis Based On Medical Knowledge Graph

With the development of Artificial Intelligence(AI)technology,methods such as knowledge graph,classification,clustering,have been considered as basic techniques for AI applications.Meanwhile,as an important part of AI applications,e-Health receives more and more attentions.Raised concern of health care management brings a significant amount of online medical services as well as websites,and therefore generates massive medical data.Integrating online professional medical knowledge into medical knowledge graph as well as designing knowledge graph based on disease diagnostic service have become challenging tasks.In this thesis,we propose a Automated Disease Diagnosis based on knowledge graph for the diagnosis of common diseases.we investigated knowledge graph construction in medical field,proposed a knowledge graph based on disease diagnostic model.The main contributions in this thesis involves two aspects:medical knowledge graph construction and disease diagnostic model design.For the medical knowledge graph construction,the work-flow of the proposed method is that,firstly,crawling medical data from professional medical websites;secondly,extracting medical knowledge by using dictionary based methods and rule based methods;thirdly,storing medical knowledge graph into Neo4J graph database by using property graph expressing methods;finally,Integrating knowledge and generating medical knowledge graph by using integrated entity alignment method.For the disease diagnostic model design,this thesis propose a relation quantification method based on the medical knowledge graph.The work-flow of the proposed disease diagnostic model is as follows.Firstly,we search the matching relation between clients’ selected feature information and the medical knowledge graph;secondly,we reason clients’ possible disease by combining the searching result and the disease-symptom relation quantification result.In this thesis,Professional doctors combine experience knowledge and clinical data to set up test suite.Theoretical analyses and evaluation results show that in the absence of physical examination indicators,the proposed knowledge graph based on disease diagnosis model achieves 91%accuracy when diagnosing common disease,with less than 0.2 seconds time overhead.The proposed methods in this thesis have a wide range of promotion value and application prospects,and have been implemented in real systems.