Research On Construction Technology Of Medical Graph Based On Electronic Medical Records

The electronic medical record is the original record in the process of diagnosis and treatment of patients,which contains a large amount of professional medical knowledge.Through the analysis of the electronic medical records,medical information closely related to patients can be mined,which can help medical staffs diagnose diseases.However,the electronic medical record belongs to unstructured text,without unified annotation standard and text structure,and mostly exists in the free-form text.Therefore,traditional natural language processing technology is difficult to extract effective medical information from it.The construction of knowledge graphs provides a good solution for the extraction and management of medical information.Aiming at the structure and language characteristics of real electronic medical record data,this thesis proposes a knowledge graph construction framework based on a deep learning model and medical knowledge base,and finally uses a graph database for storage and visualization.The specific work in this thesis is as follows.(1)In the named entity recognition module,aiming at the problem that most of the existing entity recognition methods fail to fully consider the feature of the partial component of Chinese characters and the specialty of the Chinese medical field,this thesis proposes an electronic medical record entity recognition method combining the feature of the partial component.In this thesis,we crawl Xinhua Dictionary and Baidu Chinese Dictionary to make a medical dictionary,and use the Bi LSTM model to extract the paraphernalia features of the external medical dictionary as well as to capture the correlation of the internal features,in addition,we also use the pre-trained language model Ro BERTa to extract the contextual features of the electronic medical record dataset.Finally,the feature representations extracted from both are fused and CRF is used to obtain the optimal label sequence.(2)In the relationship extraction module,this thesis proposes a relationship extraction method that combines the Cas Rel model and remote supervision to address the problem of overlapping triads in electronic medical records.The method firstly uses remote monitoring technology to automatically annotate data relationships with the help of medical knowledge base and then uses the Cas Rel model to recognize and classify entity relationship.Whereas traditional relational classifiers derive relationships between pairs of entities through head and tail entities,which is prone to the problem of relational misclassification,the Cas Rel model relationships as a function of mapping head entities to tail entities to accurately achieve relational classification,thereby solving the overlapping triad problem.(3)In this thesis,we collect and sort out the real electronic medical record data of hospitals,and conduct experiments on the real clinical electronic medical record dataset and open dataset to verify the validity and portability of the model respectively.Through in-depth analysis of entity and relationship in the electronic medical record,entity extraction and relationship extraction based on electronic medical record data are realized and successfully imported into graph database Neo4 j to realize storage and visualization of an electronic medical record knowledge graph.