基于6项肿瘤标志联合检测的3种分类模型判别肺癌的对比分析

Comparison and analysis of three classifying models for discrimination of lung cancer established by 6 tumor markers

目的联合检测6项血清肿瘤标志,建立人工神经网络(ANN)、分类回归决策树(CART)和Fisherχ2检验判别分析3种分类模型,并对肺癌进行判别,以探讨3种模型在判别肺癌中的差异。方法采用放射免疫学、分光光度法、原子吸收分光光度法等方法,测定50例正常对照、40例肺良性疾病患者及50例肺癌患者血清中癌胚抗原、胃泌素、神经元特异性烯醇化酶、唾液酸、铜锌比值(Cu/Zn)、钙(Ca)6项指标,并建立基于这6项指标的ANN、CART和Fisher判别分析3种诊断肺癌的分类模型。结果ANN、CART和Fisher判别分析模型对肺癌检出的灵敏度分别为100%、93.33%、84.00%,特异度分别为100%、100%、98.89%,对预测集正常、肺良性疾病和肺癌识别的准确度分别为91.67%、86.11%、85.00%,三模型对全部样本判别肺癌的ROC曲线下面积分别为0.964、0.953、0.812,其中ANN与CART模型ROC曲线下面积差异无显著性(P>0.05),而ANN、CART与Fisher判别分析模型ROC曲线下面积差异均有显著性(P<0.05)。结论基于6项肿瘤标志建立的ANN、CART模型判别肺癌的效果优于Fisher判别分析。

Objective To distinguish lung cancer by detecting 6 tumor markers in serum and establishing three classifying models of artificial neural networks （ANN）, decision tree （CART）, Fisher discrimination analysis, and to compare the differences among three models. Methods The levels of serum CEA, gastrin, NSE, sialic acid（SA）, Cu/ Zn, Ca in 50 healthy individuals, 40 patients with lung benign disease and 50 patients with lung cancers were detected by means of radioimmunology, spectrophotometry, atomic absorption spectrophotometry, respectively, and then developed ANN, CART and Fisher discrimination analysis models. Results The sensitivity of ANN, CART and Fisher discrimination analysis models were 100% ,93.33% ,84.00%, the specificity were 100%, 100% ,98.89%, the accuracy were 91.67 % , 86.11% , 85.00 % . The areas under receiver operating curve （ AUROC ） of ANN, CART and Fisher discrimination analysis models were 0.964,0.953,0.812, respectively. There was no significantly statistical difference between ANN and CART （P 〉 0.05）, while there were significantly statistical differences not only between Fisher discrimination analysis and ANN, but also Fisher discrimination analysis and CART（ P 〈 0.05）. Conclusion The effects of ANN, CART models established by 6 tumor markers were better than that of Fisher discrimination analysis in discrimination of lung cancer.