Research And Implementation Of Text Mining Based On Medical Record Data

With the development of information technology and advances in medicine,medical detection,monitoring and treatment technologies have developed at an unprecedented rate,and disease diagnosis methods and treatment programs are becoming more and more effective.At the same time,a large amount of data is generated during the prevention,diagnosis and treatment of diseases,so that it has to rely on information systems and large databases throughout the hospital departments to manage.The data types in the medical database are rich,including background information such as patient name,gender and medical history,various laboratory tests and measurement records accepted by the patient in the hospital,various medical treatments and nursing records made by the doctor for the patient.This is largely helpful in collecting information on symptoms,physiological conditions,and various laboratory indicators related to disease development.Comprehensive medical data can help medical staff make correct decisions about disease prevention,diagnosis and treatment.However,due to the wide variety of data types and different forms of data,the display of data is often scattered independently.It is difficult for medical personnel to quickly extract useful information from them,resulting in "data-rich,information-poor" conditions,and instead increase the number of medical staff.The burden of reading,memorizing and analyzing data.In response to the above problems,the idea of text mining is used to dig deeper into electronic medical record information.With the disease-assisted diagnosis as the entry point,a disease-assisted diagnosis model was designed and implemented.At the same time,an electronic medical record retrieval system and an auxiliary diagnosis visualization website were developed.Based on the characteristics of electronic medical records,the disease-assisted diagnosis model was designed to propose a loss function suitable for the classification characteristics of disease labels.Deriving the patient's possible diagnosis through the chief complaint and current medical history information in the electronic medical record.Establish a deep learning model,based on the 160,000 electronic medical records of the general hospital,screen out 100 common diseases in hospital based on the frequency of the disease,and carry out model training results:The top-1 accuracy rate of more than 79%of the study designed a neural network structure for the electronic medical record grammar style,including customized neural network structure,loss function and so on.