Disease Diagnosis Method Research Based On Traditional Chinese Electronic Medical Record

Disease diagnosis is the first step in clinical diagnosis and treatment.In recent years,the use of electronic medical record(EMR)data for automatic diagnosis and analysis of diseases is a hot topic in medical informatics and medical artificial intelligence research.With the wide application and promotion of the EMR in Chinese medical hospital,it is feasible to use the medical records of traditional Chinese medicine(TCM)for disease diagnosis.The main content of TCM record(inpatient record)is usually based on diagnosis and treatment process of combination of traditional Chinese and Western medicine.And in view of the difference of clinical diagnostic features between Chinese and Western medicine,it is a problem to be explored whether the related research based on EMR of Western medicine can be directly extended to the clinical environment of TCM.In this paper,the disease diagnosis method is studied by combining a certain scale of EMR and TCM domain knowledge,and a standard data set for the diagnosis of various diseases is constructed.We focused on the exploration of feature processing and representation learning,combined various classification learning models,and proposed several new diagnostic methods with certain practical value.The main research results include the following two aspects:First,according to the inclusion criteria of the diagnosis,we collected and collated a batch of EMR of TCM,and constructed a number of standard data sets of specific diseases,including chronic viral hepatitis B data set(1366 consultations),type ? Diabetes dataset(856 consultations)and cirrhosis dataset(2304 consultations).The above dataset contains many clinical features including symptoms,tongue and pulse,Chinese medicine,past history,and laboratory examination information.Based on the analysis of the importance of different types of clinical features,a weight-based feature selection method(WBFS)was proposed.The comparative experiments were performed on the set by combining with multiple classification models(such as logistic regression,SVM,ensemble learning,and Stacking classification model)in standard data.The results showed that the diagnostic method based on feature selection can achieve the effect based on the expert artificial selection method,and the diagnostic performance is significantly improved compared with the baseline method without feature selection.At the same time,the best performance is achieved when using the Stacking model(AUC value is 0.919 vs 0.676,0.922 vs 0.68,0.979 vs 0.911,on the three data sets compared with the baseline method).Secondly,we introduced a deep feature representation method based on network embedding,combined feature-representation learning with network data such as targets of traditional Chinese medicine,and proposed a disease diagnosis method based on network embedding(NEDDM).This method can expand and supplement the features of patients according to the similarity between different features,thereby enhancing the diagnostic performance.Combined with the Stacking model,the AUC values of this method on chronic viral hepatitis B,type II diabetes,and liver cirrhosis datasets reached 0.965,0.966,and 0.988,respectively.The above research shows that the network embedding based disease diagnosis method combined with the Stacking model has great advantages,and has already had practical value in predicting different diseases.It is expected to become a potential disease diagnosis method based on the EMR of TCM.