Research On The Method Of Information Extraction Based On Unstructured Medical Records

Digital information technology in the field of medical science and technology and the development of big data are constantly being promoted.Its operation mode is increasingly turning to electronic medical records.With electronic medical records as an important information resource,combined with advanced technology to further explore,and ultimately make our medical cause continue to improve.These medical medical cases record detailed records of patients during their visits,covering rich medical knowledge,which can provide patients or doctors with query and decision support,and also lay a data foundation for further medical research.However,about80% of the medical records are in an unstructured state,which is difficult to be used directly,resulting in a large number of medical resources waste.Based on the characteristics of unstructured medical records,this paper studies the method of information extraction,which is mainly divided into the following three aspects:(1)Using natural language processing technology,a method for joint extraction of entities and relations based on logical tagging scheme is proposed to complete information extraction tasks.Based on the comparison results of several commonly used sequence labeling models,it was determined that the model used in this study was Bi-LSTM-CRF based on the logical tagging scheme.With 600 medical records and 41 labels,the F1 score is 76%.After completing the joint extraction of entities and relations,the knowledge of medical records has been transformed from unstructured to structured with logical reasoning.(2)The Neo4 j graph database is used to store the extracted information,and the graph representation method is used to construct a disease-centered medical knowledge graph.While visualizing the structured information,we consider how to apply the constructed knowledge graph to the actual scene in the way of medical knowledge retrieval reasoning.(3)Model design of disease prediction and health recommendation system.After completing the information extraction unstructured medical medical records,we apply it to practical application scenarios to realize the value of informationextraction.This system is based on the newly input unstructured medical record text,which is divided into three modules: preprocessing,search and match based on the disease-centered knowledge graph and reasonableness assessment.Finally,the corresponding disease prediction and health recommendation are given.In order to evaluate the effectiveness of the IE in practical application scenarios such as disease prediction and health recommendation system,we have applied this system to grassroots clinics in nearly 1,000 townships in Fujian Province and and have collected doctors' Feedback to assess the ability of systemic disease prediction.Finally,by comparing the consistency of the disease predicted by the system with the corresponding disease in the corrected data set by professional doctors,it can be concluded that the prediction accuracy of the system can reach almost 90% in the prediction of common diseases.Thus,it proves that the system can assist doctors in guessing the disease to a certain extent.