遗传算法BP神经网络在肝硬化分期诊断中的应用

发布时间：2018-03-24 12:44

本文选题：肝硬化　切入点：BP神经网络　出处：《山西医科大学》2017年硕士论文

【摘要】：目的:将遗传算法BP神经网络模型引入到肝硬化的病例数据资料中,对肝硬化分期诊断的分类进行预测分析,利用遗传算法对BP神经网络的优化作用,弥补BP神经网络自身的不足,从而提高肝硬化分期分类预测效果。方法:通过收集2006年1月到2015年12月近10年的山西医科大学第一附属医院消化内科被诊断为肝硬化的住院患者的病例资料,针对数据的特点,分别对肝硬化分期数据分别进行Logistic回归、BP神经网络和遗传算法BP神经网络模型的建模和预测。对三种模型进行比较。选择合适的模型对肝硬化分期数据进行分类预测。结果:1、分别对Logistic回归、BP神经网络和GA-BP建模和预测,结果显示:GA-BP的ACC中位数达到90%,高于BP神经网络83.33%,远高于Logistic回归46.67%;GA-BP的TPR中位数达到95.55%,高于BP神经网络90.23%,远高于Logistic回归48.14%;GA-BP的TNR中位数达到75%,高于BP神经网络62.5%,远高于Logistic回归50%;GA-BP的PV+中位数达到95.35%,高于BP神经网络91.30%,远高于Logistic回归80%;GA-BP的PV-中位数达到77.80%,高于BP神经网络57.10%,远高于Logistic回归19.40%;GA-BP的AUC中位数达到84.4%,高于BP神经网络74.9%,远高于Logistic回归48.7%。2、比较遗传算法优化前后的预测效果,结果显示:优化前,ACC是73.33%,优化后,ACC可以达到90%,预测性能得到了改善和提升。其他五个预测评价指标也是优化后更高。结论:研究数据首先将Logistic回归与BP神经网络比较分析,BP神经网络更适合处理本次研究数据。再用遗传算法来优化BP神经网络,可以使分类预测效果提高。可以得出遗传算法优化方法的肝硬化分期分类预测效果较BP神经网络有较大的提高,具有对肝硬化分期分类预测的可行性。
[Abstract]:Objective: the genetic algorithm BP neural network model is introduced into the case data of liver cirrhosis, classification of diagnosis of liver cirrhosis were predictive analysis, optimization function based on genetic algorithm BP neural network, BP neural network to compensate for the lack of their own, so as to improve the classification effect of cirrhosis stages. Methods: collected from January 2006 through the First Affiliated Hospital in December 2015 to nearly 10 years of Shanxi Medical University was diagnosed as liver cirrhosis patients, according to the characteristics of data, respectively, on the stage of cirrhosis data Logistic regression, modeling and prediction of BP neural network and genetic algorithm BP neural network model. The comparison of three models. Choose a suitable model of staging data classification the prediction of liver cirrhosis. Results: 1, respectively, of Logistic regression, BP neural network and GA-BP modeling and prediction results: GA-BP the median of ACC reached 90%, 83.33% higher than the BP neural network, Logistic regression is much higher than 46.67%; median TPR GA-BP reached 95.55%, 90.23% higher than that of BP neural network, Logistic regression is much higher than 48.14%; median TNR GA-BP reached 75%, higher than the BP neural network is 62.5%, much higher than the median 50% Logistic regression; PV+ GA-BP reached 95.35% 91.30%, higher than the BP neural network, Logistic regression is much higher than 80%; median PV- GA-BP reached 77.80%, 57.10% higher than that of BP neural network, Logistic regression is much higher than 19.40%; median AUC GA-BP reached 84.4%, higher than the BP neural network is 74.9%, far higher than the Logistic 48.7%.2 regression, the prediction effect, compared before and after optimization of genetic algorithm optimization results show: before, ACC is 73.33%, after optimization, ACC can reach 90%, the prediction performance has been improved. The other five prediction evaluation index is optimized after higher. Conclusion: the first research data The comparative analysis of Logistic regression and BP neural network, BP neural network is more suitable for processing the research data. Then the genetic algorithm to optimize BP neural network, can make the classification prediction effect can be obtained. To improve the genetic algorithm optimization method of classification of cirrhosis stages compared with BP neural network prediction effect is greatly improved, the feasibility of liver cirrhosis stage classification and prediction.

【学位授予单位】：山西医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R575.2

【参考文献】