Research On Computer-aided Diagnosis For Chronic Kidney Disease In Traditional Chinese Medicine

Chronic renal disease,as one of the most important diseases threatening global public health,seriously affects the patients’ life.Traditional Chinese Medicine(TCM),as a medicine regulating human function,plays an important role in the diagnosis and treatment of chronic kidney diseases.Accurate diagnosis based on the information of four methods of diagnosis of patients is a key step in the process of diagnosis and treatment of TCM,by collecting symptoms,signs and other information of patients,combined with computer technology to construct the decision assistant model of TCM syndrome differentiation,to produce effective results of decision assistant model,assisting doctors to make dialectical decisions.The main research work of the thesis is as follows:Aimed at the assistant syndrome differentiation of TCM main syndromes in chronic renal diseases,combined with computer single label attribute selection technology and classifier fusion technology,syndrome differentiation model of TCM main syndromes in chronic renal diseases is proposed.For the problem that the existing single-label attribute selection algorithms have ambiguity of feature selection,a single label attribute selection algorithm RJMIM(Relation to Joint Mutual Information Maximization)based on the joint mutual information proposed,which considers the association between different attributes and categories,and also considers the interaction between different combinations of attributes and categories.In this thesis,we compare RJMIM algorithm with JMI,JMIM,DISR and IGFS algorithms,the experiment shows that the average precision value of the assistant syndrome differentiation model takes up 84.29%.Targeted for the assistant syndrome differentiation of TCM syndrome which has both deficiency and excess in chronic renal disease,combined with computer multi label learning technology,an assistant syndrome differentiation model of TCM syndrome which has deficiency and excess in chronic renal disease is suggested.In terms of the problem that the existing multi label attribute selection algorithms do not take the interaction between different attributes and different categories into account,a multi label attribute selection algorithm MDI(Max Dependency and Interaction)based on the interaction between attributes and multiple labels is put forward,when considering the association between different attributes and different categories,the algorithm also takes the interaction between different combination of attributes and different categories into consideration.In this thesis,we compare the MDI algorithm with MDMR,MLNB,MDDM algorithms,the experiment shows that after the introduction of MDI algorithm,in the parameters of the assistant syndrome differentiation model,the Hamming Loss value is 0.0852,and the index value of Micro F1 measure is 0.7342.Traditional Chinese medicine also uses the method of "classification" for differentiate symptoms."Classification" is based on analogical thinking,and it is based on clustering.Therefore,combined with computer clustering technology,the assistant syndrome differentiation model which more reflects the thinking mode of TCM is proposed.For the problem that the quality of the initial cluster center has a direct effect on the clustering in the existing clustering technology,an initial cluster center selection algorithm AWCCA(Attribute Weights Based Clustering Centres Algorithm)based on the fact that mutual information and rough set is proposed,which uses mutual information method to assign weights to different attributes and achieves better clustering effect.In this thesis,the proposed AWCCA algorithm is used to improve the performance of K-modes clustering algorithm,compared with ML-KNN,RAKEL and BP-MLL algorithm,the experiment shows after the introduction of the improved K-modes clustering algorithm,in the parameters of the assistant syndrome differentiation model,the average precision value constitutes 78.36%,the Hamming loss value is 0.1084,and the index value of Micro F1 measure is 0.7624.Directed at the assistant syndrome differentiation of chronic renal disease with integrated traditional Chinese and Western medicine,combined with multi label learning technology,an assistant syndrome differentiation model of chronic renal disease with integrated traditional Chinese and Western medicine is proposed.In view of the problem that the existing gravity-based multi label classification methods do not consider the interaction between instances of data,a classification method ITDGM(Interaction and Data Gravitation Based Model for Multi-label Classification)is proposed and combines the distance between samples and the interaction force to achieve better classification results.In this thesis,we compares ITDGM algorithm with ML-KNN,BR,RAKEL,BRKNN and BPMLL algorithms,the experiment shows that after the introduction of the proposed ITDGM algorithm,in the parameters of the assistant syndrome differentiation model,the precision value accounts for 83.33%,and the index value of tag based average F1 measure is 0.7989,and the Micro F1 measure is 0.8190.