肿瘤标志物检测结合支持向量机模型在胃癌诊断中的应用  

Application of support vector machine model for enhancing the diagnostic value of tumor markers in gastric cancer

作　　者：王辉[1] 黄钢[1] 

机构地区：[1]上海交通大学医学院附属仁济医院核医学科,200127

出　　处：《中华核医学杂志》2010年第2期87-89,共3页Chinese Journal of Nuclear Medicine

基　　金：基金项目：国家自然科学基金重点项目（30830038）;上海市重点学科建设项目（$30203）

摘　　要：目的探讨肿瘤标志物检测结合支持向量机算法在胃癌诊断中的应用价值。方法收集262例胃癌、156例胃良性病变患者及149例健康对照者的癌胚抗原（CEA）、糖类抗原（CA）125、CA19-9、甲胎蛋白（AFP）、CA50五项肿瘤标志物检测结果，选择最优核函数，应用网格搜索和交叉验证的方法优化支持向量机算法参数，建立支持向量机分类模型，进行算法测试，并与5种肿瘤标志物联合测试、Logistic回归和决策树3种常见分类算法的结果进行比较。结果对胃癌诊断数据集，联合测试、Logistic回归、决策树、支持向量机算法的分类准确性分别为46．2％、64．5％、63．9％和95．1％。与其他算法相比，支持向量机算法可以提高胃癌诊断的准确性。结论肿瘤标志物检测结合支持向量机模型在胃癌的诊断上有较大应用价值。Objective To evaluate the early diagnostic value of tumor markers for gastric cancer using support vector machine （SVM） model. Methods Subjects involved in the study consisted of 262 cases with gastric cancer, 156 cases with benign gastric diseases and 149 healthy controls. From those subjects, five tumor markers, carcinoembryonic antigen （CEA） , carbohydrate （ CA ） 125, CA19-9, alphafetoprotein （AFP） and CAS0, were assayed and collected to make the datasets. To modify SVM model to fit the diagnostic classifiers, radial basis function was adopted and kernel function was optimized and validated by grid search and cross validation. For comparative study, methods of combination tests of five markers, Logistic regression, and decision tree were also used. Results For gastric cancer, the diagnostic accuracy of the combination tests, Logistic regression, decision tree and SVM model were 46.2%, 64.5% , 63.9% and 95.1% respectively. SVM model significantly elevated the diagnostic value comparing with other three methods. Conclusion The application of SVM model is of high value in enhancing the tumor marker for the diagnosis of gastric cancer.

关 键 词：胃肿瘤 支持向量机 癌胚抗原 抗原 肿瘤相关 碳水化合物 甲胎蛋白类 诊断 

分 类 号：R735.2[医药卫生—肿瘤]