基于随机森林和误差反向传播神经网络的糖尿病性周围神经病变患病风险研究  

Risk study of diabetic peripheral neuropathy based on random forest and BP neural network

作　　者：桑袆莹 黄仕鑫 易静[1] 曾庆[1] 罗亚玲 SANG Yi-ying;HUANG Shi-xin;YI Jing;ZENG Qing;LUO Ya-ling(School of Public Health and Management,Chongqing Medical University,Chongqing 40001G China)

机构地区：[1]重庆医科大学公共卫生与管理学院,重庆400016

出　　处：《解放军医学杂志》2018年第10期877-881,共5页Medical Journal of Chinese People's Liberation Army

基　　金：国家社会科学基金(15BGL191)

摘　　要：目的建立准确判别糖尿病性周围神经病变(DPN)的分类模型,为疾病的诊断提供有价值的计算机辅助方法。方法收集重庆医科大学附属第二医院2016年1-12月确诊的2199例DPN患者及在该院健康体检的2610例体检对象的52项临床信息资料,使用R软件分别构建随机森林(RF)模型和误差反向传播(BP)神经网络模型,并比较两种诊断模型的评价指标,选择最优分类模型。结果采用随机森林和BP神经网络模型对4809例观察对象进行个体患病风险分类研究,两种模型测试样本的正确率分别为99.93%、99.58%,约登指数分别为99.85%、99.14%,ROC曲线下面积分别为0.9994、0.9959。结论两种模型的分类效果均很好,但随机森林模型在判别DPN患病风险研究中具有更高的实用性。Objective To establish an accurate classification model of identifying diabetic peripheral neuropathy (DPN),and provide valuable computer-aided diagnostic system for this diseases.Methods Fifty-two clinical information materials of 2199 DPN patients and 2610health-examined individuals with other diseases from January to December 2016in the Second Affiliated Hospital of Chongqing Medical University were collected.The random forest model and back propagation (BP)neural network model for DPN patients were constructed respectively by R software,and then compared the diagnosis model indexes to select the optimal classification model.Results The individual risk of 4809DPN patients was classified using random forest model and BP neural network model.The accuracy of the tested samples for this two fitted models were 99.93%and 99.58%,the Youden index were 99.85%and 99.14%,and the area under the ROC curve were 0.9994and 0.9959.Conclusion The classification effect of the two models are very good,while the random forest model is more practical in the study of identifying DPN patients risk.

关 键 词：糖尿病性周围神经病变 随机森林 BP神经网络 

分 类 号：R587.2[医药卫生—内分泌] R747.9[医药卫生—内科学]