Research On Machine Learning Model For Syndrome-treatment-prescription Of Chronic Glomerulonephritis In Traditional Chinese Medicine

Chronic glomerulonephrit is a common chronic kidney disease in clinical practice.The basic clinical manifestations of the disease include proteinuria,hematuria,hypertension and edema.The disease has insidious cause,long course,and slow the change of state,so it is difficult for clinical treatment.Although modern medicine has adopted such means as controlling hypertension,reducing proteinuria and lipid,the prognosis is not good.Traditional Chinese medicine has significant advantages in the treatment of the disease,such as reducing adverse drug reactions and inhibiting relapse of the disease.The rapid and large-scale growth of TCM diagnosis and treatment data makes it difficult for manual analysis.At the same time,due to the multi-source,dynamic,heterogeneous and incomplete characteristics of TCM diagnosis and treatment data,it is difficult to discover hidden knowledge from it by using existing common statistical methods.Using machine learning technology can reveal the diagnosis and treatment rules from a large number of TCM diagnosis and treatment data,and discover the effective knowledge hidden in it,so as to achieve the purpose of assisting physicians to implement rapid and accurate diagnosis and treatment.Taking chronic glomerulonephritis as the research object,we condcut the research of assistant decision for TCM syndrome differentiation,TCM treatment recommendation and TCM prescription,and design and implement TCM diagnosis and treatment assistant decision system,repectively.The main research work of this paper is as follows:1.We propose a novel topic model framework called syndrome differentiation topic model(SDTM)to dynamically characterize the process for syndrome differentiation in TCM.Firstly,we first present a novel modelling approach based on Latent Dirichlet Allocation(LDA)to discover the latent semantic relationship between symptoms and syndromes in Chinese medical records.Then,to improve interpretability of topics discovered,we label the corresponding syndromes on these topics based on Jaccard similarity coefficient.Finally,we utilize Bayesian rules to differentiate syndromes for the disease.Experimental results show the superiority of SDTM over existing topic models for syndrome differentiation,and the accuracy of syndrome differentiation is 80.14%.2.A novel binary classification algorithm,called Expected Three-view Tri-training(ETT),is proposed for TCM treatment recommendation,which can accurately predict the corresponding treatment based on patients’ symptoms.In the ETT model,we present a novel attribute partition method called Expected Three-view Partition(ETP)which can automatically partition the sample space in iterative training process.Moreover,we provide a deep metric learning method to further estimate the diversity among the three classifiers.Finally,we apply the ETT algorithm to implement TCM treatment recommendation for the disease.Experiments on twelve UCI datasets demonstrate the superiority of the proposed method compared with the state-of-the-art methods,and the average classification accuracy is as low as 86.31%.Meanwhile,the experimental results on the clinical medical record dataset show that the accuracy of ETT algorithm for TCM treatment recommendation is 81.60%,which is better than the other three related comparative algorithms.3.We propose an herb recommendation algorithm for TCM prescribing,which can effectively generate corresponding prescriptions based on patients’ symptoms.First,based on the study of TCM clinical medical records,we propose a multi-content LDA model,also known as the "Symptom-Syndrome-Herb Topic Model(SSHTM)”,which can effectively discover the corresponding relationship among symptoms,syndromes and herbs from clinical medical records.Then,based on these relationships discovered by SSHTM,a symptoms based prescription recommendation algorithm is proposed,which can help physicians quickly and accurately prescribe corresponding treatment herbs for patients’ symptoms.The experimental results show that the generalization performance of SSHTM is better than LDA model,and the symptoms based prescription recommendation algorithm is significantly better than baseline algorithm on precision,recall and F1 value,respectively.4.We designe and implement a decision-making system for TCM diagnosis and treatment.The system adopts browser/server(B/S)structure and is developed based on the Java platform Spring Boot open source lightweight framework.The system has the functions of user management,patient information collection,syndrome differentiation and treatment,prescription recommendation and generation of diagnosis and treatment results,to achieve syndrome differentiation,treatment recommendation and prescribing based on patients’ symptoms.It can effectively assist physicians for TCM diagnosis and treatment.