Research And Application Of Clinical Assistant Decision-Making Based On Medical Knowledge Graph

Knowledge graph is the basis of realizing cognitive intelligence,and it is also one of the main research hotspots in the development of knowledge engineering and artificial intelligence.Constructing medical knowledge graph and realizing the application of clinical auxiliary decision-making can effectively alleviate the shortage of medical resources,improve the efficiency of clinical diagnosis and treatment,and reduce the rate of clinical missed diagnosis and misdiagnosis.However,due to the complexity of the medical knowledge itself,the construction of medical knowledge graph and the application of clinical assistant decision-making are facing a series of challenges.This paper makes an in-depth study on the application of clinical assistant decision-making.The main research contents and innovations are as follows:(1)According to the problems of a variety of entity types with complex description,nonstandard entity name and unstructured description in Chinese electronic medical record,a new annotation method of entity recognition,alignment and structuration is proposed.This method clearly defines the scope and boundary for the entities,and synchronously realizes the entity alignment and structural annotation goals by using the JSON serialization expression.Nine predefined types of entities are extracted from Chinese electronic medical records by using deep learning algorithms Bi LSTM-CRF and Transformer.The experimental results show that the accuracy rate of entity recognition is 92.44%,the recall rate is 91.87%,and the F1 value is 0.9215;The accuracy of entity alignment and structuration is 92.78%,the recall rate is 91.63%,and the F1 value is 0.922.A novel platform for text annotation and service called n PTAS is designed and implemented for meeting the requirements of the complexity of text corpus annotation and the iterative traning of the models.n PTAS provides a onestop solution for the text data as input and the model services as the output.This research provides basic service support for the construction of medical knowledge graph and applications such as disease diagnosis,intelligent retrieval,intelligent inquiry,and clinical scientific research.(2)A construction method of medical knowledge graph and a management system for accurate medical knowledge representation are proposed and implemented.Due to the complex description of medical knowledge itself and the difficulty of knowledge representation based on triples,a series of fine-grained medical concepts are defined as well as the semantic relations with the characteristics of hierarchy,direction,and constraints at the model level of knowledge graph.The atom,compound,and combination models are proposed for the complex medical knowledge representation at the data level of knowledge graph,and the effectiveness of them are verified by cases.In order to support the construction of largescale and high-quality medical knowledge graph,a knowledge graph management system is designed and implemented.Taking the concept of disease as the core,the knowledge graph is constructed by 123 kinds of medical related concepts and 316 kinds of semantic relations.And it contains 303572 entities as well as 2113802 triples extracted from clinilcal textbooks,literatures,guidelines,and standard glossary,etc.The largescale and accurate representation of medical knowledge graph provides the core basic support for the follow-up clinical decision-making applications.(3)An analytical method of clinical test results based on medical knowledge graph is proposed.Aiming at the problems of rich and complex clinical test knowledge,diversified analysis factors of test results,and human objective factor analysis is easy to lead to missed diagnosis and misdiagnosis,a medical knowledge graph about test is constructed,and the intelligent analysis method and application of test results are realized combined with semantic model and reasoning rules.12 medical related concepts and 32 semantic relationships are clearly defined for the model of the knowledge graph;3212 entities and 14132 triples are extracted from authoritative test textbooks,standards,and literatures;A variety of impact factors about the basic information of patients,test methods,specimens,equipments and so on are combined to accurately identify the reference range of test indicators,to realize the abnormal mark of test results.Finally,the test analysis report with clinical significance and interpretation can be realized by combining with the clinical judgment rules.The research results are also effectively transformed and applied to the clinical business of medical institutions to provide diagnosis and treatment decision support for clinicians.(4)A reasoning method of disease diagnosis based on medical knowledge graph is proposed.In view of the lack of explicability of disease diagnosis based on deep learning,and the lack of diagnosis and treatment level of grass-roots doctors is prone to missed diagnosis and misdiagnosis and so on.An interpretable and accurate disease diagnosis reasoning method is realized by combining with the constructed medical knowledge graph.Taking the medical knowledge graph,basic information of the patient and clinical electronic medical record as inputs,the analysis processes of patient portrait construction,association of clinical manifestation,generation and filtering of candidate diseases,and complementation of disease knowledge are realized respectively.Based on the thinking of clinical diagnosis and treatment,the calculation method of comprehensive recommendation score of candidate diseases is realized from the two aspects of disease interpretation and diagnosis.560 real clinical medical records containing diagnostic results are used as reference standards,and the method in this paper is applied for identification and comparison.The experimental results show that the accuracy of TOP-5 is93.57%,and the accuracy of TOP-3 is 89.82%.Detailed explanation and analysis are provided for each recommended disease.The research results are also effectively transformed and mainly applied to grass-roots medical institutions to provide clinical auxiliary diagnosis services for clinicians.