基于胸腔积液肿瘤标志物的PLS-DA和ANN-MPL模型对肺癌的诊断价值分析  

PLS-DA and ANN-MPL Models to Estimate the Diagnostic Value of Four Tumor Markers in Pleural Effusion for Lung Cancer

作　　者：田刚[1] 周明术[1] 宋敏[1] 杭永伦[1] 王开正[1] 刘靳波[1] 

机构地区：[1]泸州医学院附属医院检验科,泸州646000

出　　处：《成都医学院学报》2013年第5期521-524,共4页Journal of Chengdu Medical College

基　　金：四川省卫生厅资助项目(NO:100258);泸州医学院自然科学基金资助项目(NO:12043);泸州医学院附属医院人才基金项目(NO:12277)

摘　　要：目的探讨联合检测胸腔积液中癌胚抗原(CEA)、神经元特异性烯醇化酶(NSE),细胞角蛋白19片段(CYFRA21-1)和CA125对肺癌的诊断价值。方法应用电化学发光免疫分析法测定53例肺癌和52例肺部良性疾病患者胸腔积液中4种肿瘤标志物(CEA、NSE、CYFRA21-1和CA125),结合偏最小二乘判别分析(PLS-DA)线性模型和人工神经网络多层感知(ANN-MPL)非线性模型进行建模诊断和预测分析。结果 PLS-DA模型不能完全鉴别肺癌组和对照组,具有58.5%的灵敏度、98.1%的特异性,78.1%的准确性和84.6%的预测能力。在ANN-MPL中,联合检测4种胸腔积液肿瘤标志物的受试者工作特征曲线下面积(AUC)均优于单一的肿瘤标志物,具有更高的诊断价值(AUC=0.997)。ANN-MPL诊断模型的灵敏度、特异性和准确性分别为93.9%、100.0%和96.8%。ANN-MPL预测模型的灵敏度和特异性分别为90.0%和95.5%,具有92.9%的预测准确性。结论 PLS-DA和ANN-MPL模型在肺癌的鉴别诊断中均取得了较好的效果,ANN-MPL模型更有助于肺癌的鉴别诊断和预测分析。PLS-DA和ANN-MPL模型从数据建模分析的角度证明了肿瘤标志物联合检测的重要性和临床应用价值。Objective To evaluate the diagnostic significance of four tumor markers（CEA,NSE,CYFRA21-1 and CA125）in pleural effusion for patients with lung cancer. Methods The concentration of CEA, NSE,CYFRA 21- 1 and CA125 in pleural effusion was measured by electrochemiluminescence immunoassay in 53 patients with lung cancer and 52 patients with benign pulmonary lesions. Data sets were analyzed by the line model of PLS-DA and the nonlinear model of ANN-MPL in order to search better biomarkers in the diagnosis and perception of lung cancer. Results PLS-DA model showed a moderate separation between patients with lung cancer and patients with benign pulmonary lesions. This model provided 58. 5% of sensitivity,98.1% of specificity and 78.1% of accuracy for the diagnosis of lung cancer and 84.6% power for prediction. In ANN MPL model,the AUC under the ROC based on the joint detection of four tumor markers was larger than single tumor marker, providing more powerful diagnostic importance for lung cancer （AUC： 0. 997）. ANN-MPL diagnosis model provided 93.9 % of sensitivity, 100.0 % of specificity and 96. 8% of accuracy for the diagnosis of lung cancer. ANN-MPL test model provided 90. 0% of sensitivity,95. 5% of specificity and 92. 9% power for prediction. Conclusion PLS-DA and ANN-MPL models based on the connection of four tumor markers are useful for diagnosis of lung cancer. ANN-MPL model is more powerful. Based on the model recognition technology,PLS-DA and ANN-MPL model provide a new insight showing the importance and clinical significance for the joint detection of tumor markers.

关 键 词：肺癌 肿瘤标志物 诊断 偏最小二乘判别分析 人工神经网络 

分 类 号：R734.2[医药卫生—肿瘤]