基于机器学习方法建立血清多种肿瘤标志物联合区分胃炎和胃癌的三种模型及评价  

Establishment and Evaluation of Three Models for the Distinguish Gastritis and Gastric Cancer with Multiple Serum Tumor Markers Based on Machine Learning Methods

作　　者：郝晓燕 周磊 白丛霞 刘家云 HAO Xiao-yan;ZHOU Lei;BAI Cong-xia;LIU Jia-yun(Department of Clinical Laboratory,the First Affiliated Hospital of Air Force Military Medical University,Xi’an 710032,China)

机构地区：[1]空军军医大学第一附属医院检验科,西安710032

出　　处：《现代检验医学杂志》2023年第2期124-128,共5页Journal of Modern Laboratory Medicine

摘　　要：目的基于三种机器学习方法建立多肿瘤标志物联合区分胃炎和胃癌模型。方法选取2010~2021年期间来西京医院就诊诊断为胃炎和胃癌的患者13727例,收集入组患者基本信息(年龄和性别)、甲胎蛋白(alpha-fetoprotein,AFP)、癌胚抗原(carcinoembryonic antigen,CEA)、糖链抗原19-9(carbohydrate antigen 19-9,CA19-9)及糖链抗原125(carbohydrate antigen 125,CA125)结果。采用随机森林(random forest,RF)、决策树(decision tree,DT)和K最邻近法(K-nearest neighbor,KNN)三种机器学习算法挖掘入组患者6种变量的数据,建立区分胃炎和胃癌模型。验证各模型对所有入组患者、不同年龄层的入组患者、AFP阴性入组患者的胃炎和胃癌鉴别能力,并与单肿瘤标志物鉴别能力做对比。结果利用机器学习算法构建的RF-pv6,DT-pv6和KNN-pv6模型对所有的患者诊断曲线下面积(area under the curve,AUC)均高于0.742,单肿瘤标志物AUC均低于0.644;各模型对于小于50岁患者,AUC均高于0.668,单肿瘤标志物AUC均低于0.641;各模型对于大于50岁患者,AUC均高于0.734,单肿瘤标志物AUC均低于0.647;各模型对于AFP阴性患者,AUC均高于0.731,单肿瘤标志物AUC均低于0.639。各模型在所有入组患者及其亚组中的AUC高于单肿瘤标志物的AUC。结论通过利用机器学习算法挖掘入组患者的6种特征数据建立的三种模型效能均优于单肿瘤标志物对胃炎和胃癌的鉴别能力。Objective To establish a multi-tumor marker combined distinguish model of gastritis and gastric cancer based on three machine learning methods.Methods A total of 13727 patients diagnosed with gastritis and gastric cancer in Xijing Hospital from 2010 to 2021 were selected.Collected the basic information(age,sex)of patients in each group and the detection results of alpha-fetoprotein(AFP),carcinoembryonic antigen(CEA),carbohydrate antigen 19-9(CA19-9)and carbohydrate antigen 125(CA125).After preprocessing the data,three machine learning algorithms,random forest(RF),decision tree(DT)and K-nearest neighbor(KNN)were used to mine the data of 6 variables,and distinguish gastritis and gastric cancer models were established respectively.The ability of each model to discriminate gastric cancer for all enrolled patients,enrolled patients of different age groups,and AFP-negative enrolled patients was verified,and compared with the ability of single tumor marker to discriminate gastric cancer.Results The RF-pv6,DT-pv6 and KNN-pv6 models constructed by machine learning algorithms have AUC higher than 0.742 for all gastric cancer patients,AUC higher than 0.668 for patients younger than 50 years old,AUC higher than 0.734 for patients older than 50 years old,and AUC higher than 0.731 for AFP-negative patients.The AUC of each model in the diagnosis of gastric cancer in all enrolled patients and their subgroups was higher than that of a single tumor marker.Conclusion The performance of the 3 models established by mining 6 kinds of characteristic data of the enrolled patients by using machine learning algorithm was better than that of single tumor marker in the distinguish gastritis and gastric cancer.

关 键 词：机器学习 肿瘤标志物 胃癌 分类算法 

分 类 号：R735.2[医药卫生—肿瘤] R730.43[医药卫生—临床医学]