Research On The Similarity Calculation Of Electronic Medical Records Based On Multi-level Knowledge Graph Data

Analogical reasoning is one of the thinking methods for clinicians to make diagnostic decisions,doctors integrate the patient’s symptoms,signs,auxiliary examination data,etc.,compare with the disease theory model or disease experience model,and after comparison,identification,reasoning,to obtain disease diagnosis results.Electronic Medical Record(EMR)data,as an important carrier of information in the process of patient visit,contains objective records of the whole process of patients from admission to discharge,and is the most important source of tracing patients’ conditions.In the medical big data scenario,the similarity analysis of patient medical records can provide multi-faceted predictive evidence,which is the basis for obtaining a large number of case observation knowledge and is also one of the key steps in clinical decision-macing.As the auxiliary diagnosis and treatment technology based on clinical big data has become a research hotspot,the mining and analysis of EMR data can provide assistance for computer-aided clinical computing and realize the intelligent diagnosis,treatment,prediction and prevention of patient diseases.In this study,the electronic medical record information of rheumatoid arthritis(RA)was extracted on the basis of literature research,and the deep learning model combining MC-BERT and BiLSTM-CRF was used to identify the medical entities and relationships of the electronic medical records of rheumatoid arthritis,and 9 entity types and 7 relationship categories were systematically defined.367 electronic medical records were manually annotated,and the proposed model was applied to automatically extract the information of 1619 electronic medical records,and a rheumatoid arthritis knowledge graph containing 6465 entities and 15268 entity relationships was constructed.The multi-level knowledge graph data is used to calculate the similarity of medical records,the entities,time series-relationships and events of medical record similarity calculation are designed and formulated,and the difference of similarity calculation of EMR data with different complexity and information content is compared for the first time by using the Jaccard distance measurement method,and the results show that the entity data has the highest efficiency in the calculation of medical record similarity.In the relationship formulation process,the "time series-relationship" type was designed for the first time to enhance the dynamic nature of the relationship and have a better similarity measurement effect than the relationship model without time dimension.A formal representation model of events suitable for arthritis medical record data-"TRDMME model" was designed.The proposed method is compared with the commonly used deep learning natural language processing model,and the accuracy,recall and F1 value indicators are improved compared with other models.The multi-level similarity calculation model of the medical records of four arthritis diseases,rheumatoid arthritis,osteoarthritis,psoriatic arthritis and gouty arthritis,constructed in this study,provides new ideas for the design and construction of clinical decision support tools with higher explainability and stronger universality,and has important social significance and scientific value for predicting the future health status of patients.