逆反馈人工神经网络在社区糖尿病筛查中的应用

Study on Application of Back-propagation Artificial Neural Network Model in Screening for Type 2 Diabetes Mellitus in Community Population

[目的]建立一种适合上海市社区居民经济有效的糖尿病人群筛查方法。[方法]在上海市社区人群糖尿病流行病学调查的基础上,选择目标人群进行逆反馈(back-propagation,BP)人工神经网络模型法筛查糖尿病病人。训练组和验证组用于网络模型的建立,网络输入为采用多因素Logistic回归分析筛选出的10个变量,输出变量为是否患有糖尿病。测试组资料用于验证网络模型的实用性和可靠性。[结果]当以网络输出值0.12作为判别的阈值时,该方法对人群糖尿病筛查的灵敏度和特异度分别为67.1%和为79.7%。[结论]BP人工神经网络模型对糖尿病患者具有较强的识别能力,可作为血糖检查的"前筛"工具。

[ Objective ] To establish an economical and effective screening method for diabetics in Shanghai. [ Methods ] Based on an epidemiologieal survey of diabetes mellitus in the community population in Shanghai, a back-propagation neural network（ BPNN ）was established. The data were splitted up into training group, validation group（ both utilized to train network ）, and test group（ to determine network＇s structure ）at random. The input variable of BPNN include 10 variables that were selected by the multi-factors logistic regression analysis, while the output is type 2 diabetes mellitus（ DM ）（ yes=1, no=0 ）. [ Results ] When the threshold was set up at 0,12, the sensitivity and specificity for screening type 2 DM were 67.1% and 79.7% respectively. [ Conclusion ] The models can recognize the abnormal individuals and be used as screening tool of type 2 DM.