论文部分内容阅读
[目的]通过人工神经网络建立食管癌预测模型,为大规模筛检食管癌奠定基础。[方法]对2007～2008年徐州地区食管癌患者进行病例对照研究,应用修剪算法BP神经网络、C5.0决策树、传统Logistic回归3种办法建立预测模型,并比较3种模型的预测精度。[结果]分别选择单隐层的修剪算法BP神经网络模型、修剪纯度为75%的C5.0决策树模型和Logistic回归模型建立预测模型,预测精度分别为97.82%、96.73%、94.82%,差异有统计学意义。ROC曲线下面积比较修剪算法BP神经网络模型优于C5.0决策树和Logistic回归,各曲线下面积比较差异有统计学意义(χ2=7.4405,P=0.0242)。[结论]应用修剪算法BP神经网络建立好的预测模型相比用C5.0决策树和Logistic回归建立的模型用于食管癌的初筛效果较好。
[Objective] To establish esophageal cancer prediction model by artificial neural network and lay the foundation for large-scale screening of esophageal cancer. [Methods] A case-control study was conducted on patients with esophageal cancer in Xuzhou from 2007 to 2008. The pruning algorithm BP neural network, C5.0 decision tree and traditional Logistic regression were used to establish the predictive models and the prediction accuracy of the three models was compared. [Result] BP neural network model of single hidden layer pruning algorithm was selected. The prediction model was established by pruning C5.0 decision tree model with 75% purity and Logistic regression model. The prediction accuracy was 97.82%, 96.73%, 94.82% respectively. The difference There is statistical significance. The area under ROC curve comparison pruning algorithm BP neural network model is superior to C5.0 decision tree and Logistic regression, the area under each curve is statistically significant (χ2 = 7.4405, P = 0.0242). [Conclusion] The pruning algorithm BP neural network to establish a good prediction model compared with the C5.0 decision tree and Logistic regression model established for screening of esophageal cancer is better.