Design And Implementation Of A Platform For Atlas Construction Of Digestive System Tumors

In recent years,the mortality rate of digestive system tumors in China has been increasing.How to make use of the existing clinical experience,fully explore the useful medical relationship,and prevent the growth of malignant tumors has become a problem that medical staff need to solve.Major hospitals have accumulated a large number of electronic medical records,which contain a wealth of medical knowledge,including not only the health data and examination data of patients,but also a series of treatments conducted by doctors according to the symptoms of patients,as well as the feedback after treatment.It is an important resource to carry out clinical diagnosis and treatment research and manage medical work.But the electronic medical record is unstructured medical data,there are a lot of writing habits,omission and other problems,to mine the electronic medical record information caused a great difficulty.Knowledge Graph is a semantic understanding network that integrates NLP and visualization techniques.The knowledge graph of some professional fields can provide new knowledge and new ideas to experts,increase the learning of new knowledge in this field and shorten the time for experts to make decisions.In this paper,by studying the data characteristics of EMR and the data characteristics of digestive system tumor diseases,the map framework based on digestive system tumor was constructed based on the design rules of digestive system tumor map and the design ideas of map framework.The quantitative method is used to evaluate the information extraction model,and the results show that our method is practical and scientific.The main research contents of this paper are as follows:(1)Investigate the current situation of the medical knowledge graph,conduct detailed investigation from the current situation of the knowledge graph,the research status of the medical knowledge graph,the application of the medical knowledge graph and the analysis of the research status of the digestive system tumors,and analyze the problems existing in the construction of the existing knowledge graph framework.In addition,detailed investigation was conducted on the construction of atlas framework,named entity recognition,medical relationship extraction,and unstructured data storage,which laid a good foundation for future studies on the construction of atlas framework,named entity recognition,relationship extraction,atlas storage and other aspects of digestive system tumors.(2)This chapter studies the deficiencies of the existing medical knowledge graph framework.The characteristics of digestive system tumors were investigated and analyzed.It was found that digestive system tumors were independent and single,with multiple stages.The design rules of the digestive system tumor atlas were formulated,and the design ideas of the atlas framework were introduced from the aspects of demand analysis,definition of conceptual pattern,data processing,model training,data storage and display,etc.Based on the design rules of digestive system tumor atlas,the atlas for digestive system tumor was constructed through the design idea of the atlas framework.(3)By studying the problems existing in the conceptual model construction of medical atlas and the definition of medical entities and relationships in the medical field by predecessors,combining the data characteristics of electronic medical records and the data characteristics of digestive system tumors,the digestive system tumor model was constructed,and 15 entity types and 11 relationship types of digestive system tumors were defined.A top-down approach was adopted to construct the conceptual model of digestive system tumors.Secondly,the named entity identification and relationship extraction of digestive system tumors were carried out by combining the data characteristics of electronic medical records and digestive system tumor diseases.The BILSTM-CRF model was used for named entity recognition of electronic medical record data,and compared with the BERT-BILSTM-CRF named entity recognition model.The remote supervised relationship extraction model based on Bert model was used to extract the relationship from the electronic medical records of digestive system tumors.And compared with PCNN+ATT model,RESNET model and DENSENET model.(4)Verify the construction of digestive system tumor atlas framework and the construction process of digestive system tumor atlas.Quantitative methods were used to identify the named entity of digestive system tumors and extract the relationship between digestive system tumors to verify the model.In order to verify the effectiveness of the proposed framework of tumor atlas of digestive system,the constructed tumor atlas was visualized.The results proved that the construction of digestive system tumor atlas in this paper is practical and effective.