Risk Identification Of Coronary Artery Stenosis Based On BP Neural Network And Random Forest Algorithm

Objective:Coronary heart disease is a cardiovascular disease caused by myocardial ischemia and hypoxia caused by coronary stenosis.The prognosis is complex and changeable.Coronary heart disease has become an important cause of threat to human life and health.Coronary angiography is the only method that can directly observe the morphology of coronary arteries.It is considered to be the "gold standard" for the diagnosis of coronary heart disease.However,due to the traumatic nature of coronary angiography,contraindications and postoperative complications,Because of the high cost and many other shortcomings,it is impossible to carry out large-scale population screening.Therefore,it is particularly important to construct an early risk identification model of coronary artery stenosis for patients with coronary heart disease.Based on the clinical records of coronary heart disease patients with heart failure,this study constructs an early identification model of coronary artery stenosis in patients with coronary heart disease,realizes efficient and non-invasive diagnosis of coronary artery stenosis,so as to guide clinicians and patients to choose reasonable preventive treatment and intervention measures to reduce the incidence and mortality of coronary artery stenosis.Methods:According to the inclusion and exclusion criteria of the study subjects,a total of 2926 hospitalized patients diagnosed as coronary heart disease between October 2011 and May2018 were selected from the Shanxi Provincial Cardiovascular Hospital and the First Hospital of Shanxi Medical University.The general demographic data,past history,laboratory examinations,electrocardiogram,color Doppler echocardiography,coronary angiography,medication and other medical records of patients were obtained by consulting the electronic and paper medical records in the medical record rooms of two hospitals.Chi-square test and rank-based nonparametric test were used to screen variables related to coronary artery stenosis from the above data.A stratified sampling method was used toextract three-quarters of the samples from the data of patients with Gensini score greater than or equal to 4 and patients with Gensini score less than 4(including those without coronary angiography)as training data sets,which were used to train the initial model,and the remaining quarter of the samples were used as test data sets to evaluate the comprehensive performance of each classification model.The variables selected from the medical records were used as input variables,and whether the Gensini score of coronary artery stenosis was greater than 4 as the outcome variable,logistic regression,BP neural network and random forests classification identification model were established in the training data set,respectively,the comprehensive performance was evaluated and compared by the accuracy,sensitivity,specificity,positive predictive value,negative predictive value and area under the receiver operating characteristic curve.Results:A total of 49 independent variables related to coronary artery stenosis,including arrhythmia,hypertension,hemoglobin,platelets,etc.,were initially screened from 147 variables by single factor test(chi-square test and rank-based nonparametric test).The variables obtained by the above-mentioned single factor test were further screened by the AUC-based random forests independent variable screening method,and finally 36 variables entered the training of the final three models.The 36 variables obtained in the above process are used as input variables to train the initial model of logistic regression,BP neural network and random forests.The results of logistic regression model in test data set: sensitivity was 75.76%,specificity was 72.95%,accuracy was 74.05%,positive predictive value was 73.95%,negative predictive value was 72.07%,AUC value was0.7399.Before training the BP neural network,firstly,through the simulation experiment,the number of hidden layers is determined to be 25,and the neural network model with the model structure of 36-25-1 is constructed.The result of the BP neural network model in the test data set: the sensitivity was 74.30%,the specificity was 70.00%,the accuracy was72.30%,the positive predictive value was 75.05%,the negative predictive value was69.18%,and the AUC value was 0.7231.Before training random forests model,the model parameters mtry and ntree are selected.When mtry is set to 3 and ntree is set to 1000,the model performance achieves the best.In the test data set,the model effect of random forests model is: the sensitivity was 93.70%,the specificity was 62.97%,the accuracy was79.49%,the positive predictive value was 74.58%,the negative predictive value was89.39%,the AUC value was 0.7522.Conclusions:Random forest model has the best comprehensive performance in the identification of coronary artery stenosis degree,which can diagnose the coronary artery stenosis in the early stage of coronary heart disease,and it makes the incidence of coronary artery stenosis possible.It is estimated to be more accurate,and it provides clinical doctors and patients with more accurate and efficient opinions and suggestions,which is very important for extending the life cycle of patients and improving their quality of life.