Construction And Application Of Multi-decision Syndrome Differentiation Model In TCM Intelligent Auxiliary Diagnosis And Treatment System

Objective:In view of the current deficiencies in the construction and application of intelligent syndrome differentiation,this study proposes the research idea of building a multi-decision syndrome differentiation model.Taking the extraction of text information,optimization of medical record data structure and model selection and joint construction as an important part of the construction of multiple decision model,the multiple decision dialectic model is constructed by adopting the strategy of the Krem correlation coefficient combined with the confidence weighted voting algorithm.It aims to provide reference for improving the accuracy of artificial intelligence syndrome differentiation model and better guide clinical syndrome differentiation.Method:Through the collection and sorting of the first to fifth batch of clinical documents of famous and old Chinese medicine in the treatment of coronary heart disease contained in the Chinese Academic Journals Fulltext Database(CNKI),Wanfang Database,and Vip Chinese Science and Technology Journals Full-text Database from the establishment of the database to December 2022,we screened out the modern and famous Chinese medicine cases of traditional Chinese medicine in the treatment of coronary heart disease,and created a table database,the database includes the literature source,patient’s age,gender,symptoms,physical signs,tongue picture,pulse picture and syndrome type.The literature is sorted into data information that can be recognized by the computer by means of double input,unify and standardize the terms such as symptoms and syndrome types of traditional Chinese medicine with similar expressions to ensure the maximum recognition and application of information.Using statistical methods such as frequency analysis and chi-square test,this paper analyzes the basic data of famous old Chinese medicine in treating coronary heart disease.SPSS Modeler18.0 software is used to model and analyze the sorted database of coronary heart disease.First,the automatic classifier is used to model all models of each syndrome type of coronary heart disease.According to the accuracy rate of each model and the area under the ROC curve,the model with better effect on the differentiation of coronary heart disease is selected.Select the confidence weighted voting strategy to integrate and analyze the selected models,and obtain the dialectic accuracy of weighted voting algorithm.Clem coefficient feature selection method was used to compare the variation of syndrome differentiation accuracy of selected models before and after feature selection.Result:A total of 405 eligible TCM patients with coronary heart disease were collected.The top five syndrome types with the largest proportion of cases were phlegm and blood stasis syndrome,qi and yin deficiency syndrome,qi deficiency and blood stasis syndrome,yang deficiency and blood stasis syndrome,and qi stagnation and blood stasis syndrome.Among them,there are 240 male patients,accounting for 59.26% of the total number of cases,165 female patients,accounting for 40.74% of the total number of cases,and the proportion of male and female patients is 1.45:1.The average age of the patients was 62.60 ± 9.80 years old,of which the number of patients in the 61-70 age group was the largest,with 154 patients,accounting for38.02%.The most frequent symptoms were chest pain,followed by chest tightness,fatigue,shortness of breath,palpitation and poor sleep;The tongue with the highest frequency is dark red,followed by red,fat,dark and purple;The tongue coating with the highest frequency is thin and white coating,followed by white and greasy coating,white coating,thin and yellow coating,yellow and greasy coating;The pulse with the highest frequency is the pulse with thin pulse,followed by pulse with thin pulse,pulse with smooth pulse,pulse with thin pulse,pulse with deep pulse,and pulse with weak pulse.SPSS Modeler18.0 software was used to model and analyze the five syndrome types of coronary heart disease based on the automatic classifier node,and the modeling results of each syndrome type were summarized and analyzed according to the order of the accuracy of the model differentiation.C5,KNN algorithm,SVM,CHAID,C&R tree and BP neural network ranked the top six in the accuracy of syndrome differentiation of the model of qi deficiency and blood stasis;BP neural network,KNN algorithm,Quest,C5,CHAID and SVM ranked the top six in the accuracy of syndrome differentiation of the model of deficiency of qi and yin;The accuracy of phlegm and blood stasis syndrome differentiation model ranked the top six in BP neural network,C5,CHAID,Quest,C&R tree,SVM;BP neural network,KNN algorithm,C5,SVM,CHAID and Quest ranked the top six in the accuracy of syndrome differentiation of the model of qi stagnation and blood stasis;The accuracy of the model of Yang-deficiency and bloodstasis syndrome is ranked in the top six by BP neural network,CHAID,C5,Quest,KNN algorithm and SVM.Based on the comprehensive analysis of the area under the ROC curve,the BP neural network,SVM,CHAID and C5 models are selected for the integrated modeling of coronary heart disease.The confidence weighted voting strategy is selected for integrated modeling.The accuracy of the weighted voting algorithm for the syndrome of qi deficiency and blood stasis is 85.61%,the accuracy of the weighted voting algorithm for the syndrome of qi deficiency and blood stasis is 92.42%,the accuracy of the weighted voting algorithm for the syndrome of phlegm and blood stasis is 88.64%,the accuracy of the weighted voting algorithm for the syndrome of qi stagnation and blood stasis is 97.73%,and the accuracy of the weighted voting algorithm for the syndrome of yang deficiency and blood stasis is 90.15%.Sorting the accuracy rate of model differentiation before and after integration,the result is: Qi deficiency and blood stasis syndrome: weighted voting=C5>SVM>CHAID>BP neural network>C5>CHAID>SVM;Phlegm-stasis syndrome:BP neural network>weighted voting>C5=CHAID>SVM;Qi stagnation and blood stasis syndrome: weighted voting>BP neural network>C5=CHAID=SVM;Yang deficiency and blood stasis syndrome: BP neural network=CHAID>weighted voting>C5>SVM.The weighted voting algorithm ranks the first in the effect of syndrome differentiation of qi deficiency and blood stasis syndrome,qi and yin deficiency syndrome and qi stagnation and blood stasis syndrome,and the second and third in the effect of syndrome differentiation of phlegm and blood stasis syndrome and yang deficiency and blood stasis syndrome,respectively.The results suggest that the integrated learning method based on weighted voting strategy can effectively improve the accuracy of syndrome differentiation to a certain extent.Applying the Cramer’s V to feature selection processing,the accuracy of the weighted voting algorithm for the syndrome of qi deficiency and blood stasis increased to 87.12%,the accuracy of the weighted voting algorithm for the syndrome of qi deficiency and blood stasis increased to 93.64%,the accuracy of the weighted voting algorithm for the syndrome of phlegm and blood stasis increased to 90.12%,the accuracy of the weighted voting algorithm for the syndrome of qi stagnation and blood stasis decreased to 93.94%,and the accuracy of the weighted voting algorithm for the syndrome of yang deficiency and blood stasis increased to 94.70%.In addition to the decline in the accuracy of the weighted voting algorithm for the syndrome of qi stagnation and blood stasis,the accuracy of the weighted voting algorithm for the other syndrome types increased.Conclusion:The method based on weighted voting combined with multi-model integrated modeling proposed in this study can obtain ideal syndrome differentiation results in TCM syndrome differentiation of chest pain and heartache,and the feature selection method based on correlation analysis of Clem coefficient can further improve the accuracy of syndrome differentiation.This shows that the idea of multi-decision model construction based on feature selection at the data level and integrated modeling at the model level can be applied to the syndrome differentiation method of the intelligent diagnosis and treatment system of traditional Chinese medicine.