论文部分内容阅读
目的探讨BP神经网络筛选疾病相关因素及构建疾病预测模型的作用,并与Logistic回归模型的分析结果进行对比,为更加准确地运用神经网络方法解决医学实际问题提供科学依据。方法利用MATLAB软件中的神经网络工具箱,建立代谢综合征相关影响因素的BP神经网络模型,通过计算平均影响值对影响因素进行筛选,依据ROC曲线下面积对比BP神经网络与Logistic回归分析所构建的疾病预测模型的效果。结果利用BP神经网络所构建的疾病预测模型,通过平均影响值算法筛选变量后的预测效果要好于未筛选的效果,且通过平均影响值算法所筛选的影响因素与Logistic回归分析基本一致,两者预测效果差异无统计学意义(AUCBP=0.837,AUCLogistic=0.841,u=0.3310,P=0.7406)。结论运用BP神经网络的平均影响值算法可实现对疾病相关因素的筛选及构建疾病预测模型,可在流行病学病因探索的研究中发挥与Logistic回归分析同样的作用。
OBJECTIVE: To investigate the related factors of disease screening and the construction of disease prediction model by BP neural network and to compare with the results of Logistic regression model, so as to provide a scientific basis for more accurately using neural network method to solve practical medical problems. Methods The neural network toolbox of MATLAB software was used to establish the BP neural network model of influencing factors of metabolic syndrome. The influencing factors were screened by calculating the average influence value. According to the area under ROC curve comparing with BP neural network and Logistic regression analysis The effect of the disease prediction model. Results The disease prediction model constructed by BP neural network was better than the unselected ones by screening the variables by the average influence value algorithm and the influencing factors by the average influence value algorithm were basically consistent with Logistic regression analysis. There was no significant difference in predictive value (AUCBP = 0.837, AUCLogistic = 0.841, u = 0.3310, P = 0.7406). Conclusion The average influence value algorithm of BP neural network can be used to screen the disease-related factors and construct the disease prediction model. It can play the same role as the Logistic regression analysis in the research of epidemiological etiology.