Research On Machine Learning Models For Diagnosis And Treatment Of Chronic Glomerulonephritis In Traditional Chinese Medicine

Chronic glomerulonephritis is a kind of chronic kidney disease.The disease has the characteristics of concealment,complexity and long-term periodicity,and its pathogenesis is not clear.Traditional Chinese Medicine（TCM）,as traditional medicine of China,can comprehensively regulate human body function and has unique advantages in treating chronic glomerulonephritis.At present,due to the heterogeneous,nonlinear and multi-source characteristics of TCM diagnosis and treatment data,it is not easy to find hidden knowledge and rules,and it is difficult to make diagnosis and treatment decisions quickly and accurately.In this thesis,machine learning technology is used as a means to establish an auxiliary diagnosis and treatment model of TCM,to learn and mine the knowledge and rules of TCM diagnosis and treatment,so as to help doctors make quick and accurate diagnosis and treatment decisions.In this thesis,chronic glomerulonephritis is taken as the research object,the auxiliary diagnosis and treatment model of TCM syndrome differentiation,TCM treatment determination,and TCM prescription organization is established.TCM auxiliary diagnosis and treatment system is designed and implemented.The main work of the thesis is as follows:1.Aiming at the syndrome differentiation of TCM main symptom,a Dynamic Relevance based Feature Selection（DRFS）algorithm is proposed to measure interactivity of symptom and construct TCM main symptom syndrome differentiation model to realize TCM syndrome differentiation of main symptom,with an accuracy of 83.38%.Aiming at the syndrome differentiation of deficiency and excess,a label specific feature algorithm named Gravitation Model based Label spec Ific Fea Tures（GMLIFT）is proposed,which uses the symptom and syndrome information of similar patient,to construct TCM deficiency and excess syndrome differentiation model to realize the syndrome differentiation of deficiency and excess,with the Hamming Loss is 5.23%.2.Aiming at the treatment determination of TCM,a graph attention network named Degree and Entropy based Graph ATtention network（DEGAT）is proposed,which uses the information of similar patient and constructs the relationship between patient,to construct TCM treatment determination model to realize the prediction of TCM treatment,with an accuracy of 82.15%.3.Aiming at the drug selection of TCM prescription organization,a multi-label classification algorithm named Feature Information and Label Relationship（FILR）is proposed,which considers the relationship between symptom and drug,measures the correlation between drug.Aiming at the dosage usage of TCM prescription organization,a dose recommendation algorithm named Mean Value and Pharmacopoeia based Dosage Recommendation Algorithm（MVPDRA）is proposed,which considers medication rule of doctor and the dosage of pharmacopoeia.FILR algorithm and MVPDRA algorithm are proposed to form TCM prescription organization model to realize drug judgment and dose prediction with 82.90% and 80.00% accuracy,respectively.4.A TCM auxiliary diagnosis and treatment system is designed and built.The system is based on Spring Boot,with B/S mode,and My SQL is used as the database.The system is programmed in Java.The system has the functions of syndrome identification,treatment prediction and prescription recommendation,to assist doctor in making diagnosis and treatment decisions.