中文摘要：

目的：应用决策树技术联合肿瘤标志蛋白芯片建立基于“优选肿瘤标志群”的决策树模型，实现对肺癌的快速诊断。方法：运用肿瘤标志定量检测试剂盒测定201例肺部良性疾病及199例肺癌患者血清中9项肿瘤标志[癌胚抗原、糖原类抗原19-9（ CA199）、神经元特异性烯醇化酶、CA242、铁蛋白、CA125、甲胎蛋白、人生长激素和CA153]水平，应用logistic回归对肿瘤标志进行筛选以获得“优选肿瘤标志群”，分别于筛选前后建立决策树模型和Fisher判别分析模型。结果：肺癌组9项血清肿瘤标志水平均高于肺良性疾病组（P＜0．05）。筛选前基于9项肿瘤标志分别建立的Fisher判别分析模型、决策树模型和筛选后基于6项肿瘤标志建立的Fisher判别分析模型、决策树模型，其预测准确度分别为86．0％、92．5％、84．5％、91．5％。筛选前和筛选后决策树模型ROC曲线的AUC分别为0．925和0．915，均高于Fisher判别分析的0．860和0．845（Z＝4．462和4．575，P均＜0．01）；但决策树模型和Fisher判别分析筛选前后自身相比，差异均无统计学意义（Z＝1．914和1．074，P均＞0．05）。结论：基于6项肿瘤标志建立的决策树模型诊断肺癌的效果优于Fisher判别分析。

英文摘要：

Aim：To establish decision tree model based on filtered biomarkers to achieve rapid diagnosis of lung canc -er.Methods：The serum levels of 9 tumor markers （CEA,CA199,NSE,CA242,Ferritin,CA125,AFP,HGH and CA153） in 199 patients with lung cancer and 201 patients with benign pulmonary lesion were measured by multiple tumor marker protein biochip, and the models of C5.0 and Fisher discrimination analysis were developed based on the tumor markers be-fore and after being filtered by logistic regression .Results：The serum levels of the 9 tumor markers in patients with lung cancer were significantly higher than those in patients with benign pulmonary lesion （ P＆lt;0 .05 ） .The accuracies of Fisher discrimination analysis and C5.0 models based on 9 tumor markers and 6 tumor markers filtered by logistic regression were 86.0%,92.5%,84.5% and 91.5%, respectively.The area under receiver operating curve （AUC） of C5.0 model was higher than that of Fisher discrimination analysis in both of 9 tumor markers model and 6 tumor markers model（Z=4.462 and 4.575,P＆lt;0.01）.However, there was no significant difference in AUC between before and after screening in both models（Z=1.914 and 1.074,P＆gt;0.05）.Conclusion：The effect of the model of C5.0 is better than Fisher discrimina-tion analysis in diagnosis of lung cancer especially based on the tumor markers screened by logistic regression .