Research On Decision Support System For TCM Formulation In Chronic Glomerulonephritis

Chronic glomerulonephritis is a type of chronic kidney disease.The symptoms of proteinuria,hematuria,hypertension and edema are the clinical manifestation of the chronic glomerulonephritis.In recent years,the method that using traditional Chinese formula to cure chronic kidney disease has been extensively used in clinical practice.It is the key step that personal formulation for patients in diagnosis and treatment process of Traditional Chinese Medicine(TCM).Through collecting the symptoms of patients by four diagnostic methods,doctors distinguish syndrome based on the knowledge of Chinese medicine and give the appropriate formula.Finally,we implement the work of helping doctors to make medical decision.In this thesis,according to the diagnosis and treatment process of TCM,we get clear the key problems in TCM formulation firstly.Secondly,multiple models are proposed for solving these fundamental problems step by step.Finally these models are combined to build a decision support system for Chinese medicine formulation.It helps doctors make medical decision.Specific work is as follows:1.The function of decision support system for Chinese medicine formulation includes the syndrome distinguishing,the symptom-herb relationship mining and the Chinese medicine formula determinating.In order to distinguish the syndrome,association rule is utilized according to the diagnosis and treatment process of TCM.Then in order to find the symptom-herb relationship,two methods genetic algorithm and Chi square test are used.Genetic algorithm mines the symptom-herb relationship,and its fitness function is designed by the many-to-many relationship between symptoms and herbs in medical records.According to the frequency of herbs and symptoms in knowledge base,Chi square test also mines the symptom-herb relationship.Compared the results of two methods,the Chi square test costs average 3.11 seconds and the genetic algorithm costs average 12.35 seconds.Finally in order to determine the Chinese medicine formula,two methods KNN and SVM are used.The KNN algorithm determines the formula by weight voting strategy,and the SVM algorithm determines the formula after dealing the unbalanced dataset.Compared the results of two methods,the F1-score of SVM is 71.3% by 10-fold cross validation,and SVM has the higher accuracy than KNN.2.The valuation and recommendation of decision support system for Chinese medicine formulation includes the evaluation of Chinese formula and the feedback of symptom-syndrome-herb path after modifying by doctors.In order to evaluate the completed formula,the evaluation indicator is used in this thesis.The evaluation indicator contains the efficacy and attribute of formula.The efficacy of formula is calculated by the regularized BP neural network.And the attribute of formula is calculated by the association rule.Finally,the weight voting is used for the evaluation of the formula.Then in order to get the feedback information of the path of symptom-syndrome-herb after modifying,gradient ascent is utilized in this thesis.3.The implementation and design of the system archetype includes two function modules.The decision support for Chinese medicine formulation and its valuation and recommendation is implemented in the archetype by C# and WPF.