Prediction Of Hepatic Cirrhosis With Upper Gastrointestinal Bleeding Based On Particle Swarm Optimization BP Neural Network

Objective:The particle swarm optimization algorithm was used to train the parameters and structure of BP neural network,combining the advantages of particle swarm optimization algorithm and BP neural network to improve the model performance.At the same time,the established models were applied to the prediction of liver cirrhosis complicated with upper gastrointestinal hemorrhage,and provided a simple,noninvasive and accurate diagnosis method for patients with cirrhosis.Methods:The medical records of cirrhosis patients who were admitted to the Second Hospital in Shanxi Medical University,Department of Gastroenterology during January 2007 to June 2017.Using chi-square test and t test screening of variables associated with liver cirrhosis complicated with upper gastrointestinal bleeding,and ruled out for diagnosing liver cirrhosis complicated with upper gastrointestinal hemorrhage clinical variables.The data was randomly divided into training set and testing set according to the proportion of3:1,repeat to choose 100 times.The Logistic regression model,the BP neural network model and particle swarm optimization BP neural network were established in the training set,performance evaluation model based in the testing set,record the results of each test set and evaluation,evaluation index including true positive rate,true negative rate,accuracy,positive predictive value,negative predictive value,and the area under theROC curve,the central tendency of evaluation indicators was described by using median,the dispersion of evaluation indicators was described by using upper and lower quartile.Results:Through the selection of variables,there were 56 variables to enter the model as independent variables.The prediction results of the Logistic regression model in the test set were as follows: the median of true positive rates was 0.4910,the median of true negative rate was 0.4992,the median of accuracy was 0.4973,the median of positive predictive value was 0.2136,the median of negative predictive value was 0.4542,the median of the area under the ROC curve was 0.4990.The prediction results of the BP neural network model in the test set were as follows: the median of true positive rates was0.8748,the median of true negative rate was 0.5547,the median of accuracy was0.8058,the median of positive predictive value was 0.8763,the median of negative predictive value was 0.6909,the median of the area under the ROC curve was 0.7139.The prediction results of the particle swarm optimization BP neural network model in the test set were as follows: the median of true positive rates was 0.8933,the median of true negative rate was 0.6137,the median of accuracy was 0.8292,the median of positive predictive value was 0.8926,the median of negative predictive value was 0.7966,the median of the area under the ROC curve was 0.7522.By comparing the evaluation indexes,the particle swarm optimization BP neural network model was the best in predicting the upper gastrointestinal bleeding in cirrhosis.Conclusion:The performance of BP neural network model is better than that of Logistic regression model,and the particle swarm optimization BP neural network model is better than that of BP neural network model.According to the medical records of patients with cirrhosis implementation for the forecast of whether patients with liver cirrhosis complicated with upper gastrointestinal bleeding,provide the basis for early detection ofupper gastrointestinal tract is used to,so as to further intervention and preventive treatment.