人工神经网络在肝硬化食管大静脉曲张预测中的应用

The Use of Artificial Neural Network in Predicting Large Esophageal Varices in Patients with Cirrhosis

目的利用人工神经网络模型(ANN模型)预测肝硬化食管大静脉曲张。方法共纳入309例肝硬化患者,根据胃镜结果将患者分为食管无或小静脉曲张组和食管大静脉曲张组,记录年龄、生化指标、脾脏厚度等14个参数。先经单变量分析筛选出与食管大静脉曲张相关的指标,后用这些指标构建ANN模型。结果单变量分析显示血小板计数、凝血酶原时间、脾脏厚度、腹腔积液、门静脉宽度与食管大静脉曲张相关.以这5指标构建ANN模型预测肝硬化食管大静脉曲张的敏感度为94.27%,特异性达75.00%,诊断的准确率达到84.79%。结论 ANN模型在非侵入性预测肝硬化食管大静脉曲张方面有一定临床意义。

Objective To use artificial neural network （ANN） to predict large esophageal varices in patients with cirrhosis. Meth- ods Totally 309 patients with cirrhosis were included and subdivided into 2 groups （patients with no or small varices, patients with large varices）. Age, biochemical parameters, spleen width,et al were recorded. The candidates for input nodes of the ANN were assessed by univariate analysis. An ANN was constructed to predict the large esophageal varices. Results Univariate analysis showed that platelet count, prothrombin time, spleen width, ascites, portal vein diameter were relevant to the presence of large esophageal varices, and an ANN model developed with the 5 variables had a sensitivity of 94.27% , specificity of 75.00% and a diagnostic accuracy of 84.79% for the prediction of large esophageal varices. Conclusions An ANN may be a noninvasive diagnostic mean for predicting presence of large esophageal varices in patients with cirrhosis.