Design And Implementation Of Assisted Inquiry System Based On Electronic Medical Record

The continuous accumulation of electronic medical record(EMR)data in the medical industry provides a large amount of data support for the research and design of smart medical systems.Therefore,the development of an assisted inquiry system based on medical record data can help doctors conduct interrogation and provide effective reference opinions for diagnosis.Based on the actual inquiry workflow of general outpatient clinics in this hospital,the paper uses machine learning and natural language processing methods to construct an inquiry model to reason the doctor’s diagnosis process.The inquiry model simulates the dynamic interaction between doctors and patients,and divides the medical records into finer granularity(symptoms),so as to achieve the needs of assisted inquiry.The main work of this article is:(1)Realize the structured operation of EMRDuring the process of Interrogation,the doctor needs to repeatedly ask the patient if they have other symptoms to make the final diagnosis.Therefore,the assisted inquiry system expects to recommend the symptom words that the doctor want to ask.Recommendations are symptom-related information.Extracting the symptoms elements in the electronic medical record text is the basic work.This article completes the electronic medical record structured operation by defining medical entities,data annotations,and medical entity identification.A total of 8 entities are defined in this article,and more than 10,000 pieces of medical record data are marked.Finally,by using the BILSTM-CRF model and rule recognition strategy,the accuracy rate is 0.938.(2)Design and Implementation of inquiry model based on electronic medical record dataThe model includes two parts:symptoms recommendation and disease prediction.According to the strategy of question first and then diagnosis an inquiry model based on N-Gram and LSTM text classification was implemented.According to the strategy of first diagnosis and then question,the inquiry model based on decision tree and LSTM text classification and the inquiry model based on decision tree and reinforcement learning are realized.By selecting the medical records of ten diseases with a large number of medical records in the four departments of Neurology,Gastroenterology,Cardiology and Pediatrics as the training set of the model,the three models are compared and analyzed from the perspective of disease diagnosis accuracy and dialogue rounds.Among them,the inquiry model based on decision tree and reinforcement learning is superior to the other two models.When the number of dialogue rounds is 6,the diagnostic accuracy rate is 0.679.(3)Development of interactive website based on three inquiry modelsIn order to use the inquiry model in practice,this article develops an assisted inquiry system.The system provides medical record management,symptoms recommendation,and disease prediction.The symptom recommendation module is used for interactive assisted The symptom recommendation module is used for interactive assisted interrogation,and the disease prediction module is used for direct auxiliary diagnosis.Through the interactive website,users can switch between three consultation models to get different recommended symptom words and predict diseases for the next inquiry reference.