摘要
目的以实验室指标建立诊断2型糖尿病肾病的GA-BP神经网络模型并评价其诊断性能。方法收集2016年1-12月重庆、贵州、四川的5所三级医院确诊为2型糖尿病肾病的患者,采用SPSS 19.0和MATLAB 2014a软件对患者的89项信息进行单因素分析,提取有统计学意义的变量,以此分别构建logistic回归、BP神经网络和GA-BP神经网络模型,对比三种模型的诊断性能。结果共477例2型糖尿病肾病患者和449例对照纳入模型分析,单因素分析结果显示42项信息差异有统计学意义。Logistic回归分类模型有12个变量纳入最佳回归方程。BP和GA-BP神经网络的输入层、隐含层和输出层分别有42、15和1个节点。Logistic回归分类模型、BP神经网络模型和GA-BP神经网络模型(训练集、测试集)的约登指数分别为0.76、0.87、0.84和0.81,对数据集的分类准确率分别为88.12%、93.41%、92.09%和90.48%,ROC曲线下面积分别为0.95、0.98、0.97和0.98。结论 GA-BP神经网络模型对2型糖尿病肾病有较好的辅助诊断价值,但仍需进一步临床检验。
Objective To develop a genetic algorithm back propagation(GA-BP) neural network model based on laboratory tests for diagnosis of type 2 diabetic nephropathy, and evaluate its diagnostic performance. Methods Data of patients with diagnosed type 2 diabetic nephropathy were collected in five tertiary hospitals of Chongqing, Guizhou and Sichuan provinces from Jan. 2016 to Dec. 2016. Totally 89 items of information were analyzed by univariate analysis to identify the significant variables with SPSS 19.0 and MATLAB 2014 a. The logistic regression model, BP and GA-BP neural network model were established based on the correlated information, and the diagnostic performance of the three models was compared. Results A total of 477 patients with type 2 diabetic nephropathy and 449 subjects as control were enrolled in the model analysis. Univariate analysis showed that 42 variables had significant difference. Logistic regression analysis showed that 12 variables were included in the optimal regression equation. BP and GA-BP neural network had 42 input layer nodes, 15 hidden layer nodes and 1 output layer nodes. For logistic regression analysis, BP and GA-BP neural network(Training set and Test set), the Youden's indexes were 0.76, 0.87, 0.84 and 0.81, respectively; the classification accuracy for data sets were 88.12%, 93.41%, 92.09% and 90.48%, respectively; and the area under curve(AUC) were 0.95, 0.98, 0.97 and 0.98, respectively. Conclusions The GA-BP neural network model has important auxiliary diagnostic value for diagnosis of type 2 diabetic nephropathy. However, further clinic validation is necessary for all the conclusions.
作者
黄仕鑫
杨艳艳
罗亚玲
陈天瑶
HUANG Shi-xin;YANG Yan-yan;LUO Ya-ling;CHEN Tian-yao(College of Medical Informatics, Chongqing Medical University, Chongqing 400016, China)
出处
《解放军医学杂志》
CAS
CSCD
北大核心
2018年第6期483-489,共7页
Medical Journal of Chinese People's Liberation Army
基金
国家社会科学基金项目(15BGL191)
关键词
BP神经网络
遗传算法
诊断模型
2型糖尿病肾病
认知模式
BP neural network
genetic algorithm
diagnostic model
type 2 diabetic nephropathy
cognitive model