预测ER弱阳性乳腺癌状态的机器学习模型的建立及验证  

Development and validation of a machine learning model for predicting the status of ER-low-positive breast cancer

作　　者：徐梓航 牛淑瑶 沈荣波 贾占莉 商久妍 王新乐[4] 张硕[4] 刘月平[2] XU Zi-hang;NIU Shu-yao;SHEN Rong-bo;JIA Zhan-li;SHANG Jiu-yan;WANG Xin-le;ZHANG Shuo;LIU Yue-ping(Basic Medical College of Hebei Medical University,Shijiazhuang 050011,China;Department of Pathology,the Fourth Hospital of Hebei Medical University,Shijiazhuang 050011,China;Tencent AI Lab,Shenzhen 518000,China;Department of Breast Center,the Fourth Hospital of Hebei Medical University,Shijiazhuang 050011,China)

机构地区：[1]河北医科大学基础医学院,石家庄050011 [2]河北医科大学第四医院病理科,石家庄050011 [3]腾讯人工智能实验室,深圳518000 [4]河北医科大学第四医院乳腺中心,石家庄050011

出　　处：《临床与实验病理学杂志》2023年第7期782-787,共6页Chinese Journal of Clinical and Experimental Pathology

摘　　要：目的探讨利用机器学习算法预测ER弱阳性乳腺癌的状态。方法收集710例原发性浸润性乳腺癌,其中139例ER阴性(<1%)和311例ER阳性(>10%)乳腺癌作为训练队列,260例ER弱阳性(1%~10%)乳腺癌作为测试队列。深度学习分割模型(LinkNet)用于分割并提取肿瘤细胞的形态特征。基于朴素贝叶斯机器学习算法,利用从训练队列中提取的12个临床病理特征和14个形态特征开发机器学习预测模型,并进行内部验证。利用ROC曲线的曲线下面积(AUC)反映预测模型的性能。利用预测模型对测试队列进行ER状态预测。对比分析两组的临床病理特征、ESR1 mRNA的表达水平和预后。结果ER阴性与ER阳性乳腺癌在组织学类型(P=0.01)、淋巴结转移(P=0.02)、组织学分级(P<0.001)、PR(P<0.001)、HER2(P<0.001)和Ki-67(P<0.001)表达差异有显著性。基于朴素贝叶斯机器学习算法构建预测模型,5倍交叉验证显示,在训练队列中预测模型对ER状态的预测性能优异(AUC=0.91±0.03)。ER状态预测结果显示,260例ER弱阳性乳腺癌中206例(79.23%)被划分为阴性组,54例(20.77%)被划分为阳性组。与ER阳性组相比,ER阴性组组织学分级更高、Ki-67高表达、ESR1 mRNA表达水平低,内分泌治疗获益更少,患者预后更差。结论机器学习模型能够较为精准地对乳腺癌ER表达状态进行预测,为进一步明确ER弱阳性乳腺癌的状态提供了新视角,协助临床医师做出更为精准的治疗决策。Purpose To predict the status of ER-low-positive breast cancer by using a machine-learning algorithm with clinicopathological characteristics.Methods A total of 710 primary invasive breast cancer patients’data were included.The deep learning segmentation model was used to segment the tumor cells.Based on the machine learning algorithm,data from ER-negative(<1%)breast cancer(n=139)and ER-positive(>10%)breast cancer(n=311)with 12 clinicopathological characteristics and 14 morphological features were used to develop the machine learning model and to perform internal validation.The area under the curve(AUC)of the receiver operating characteristic(ROC)curve was used to reflect the performance of the predictive model.260 patients with ER-low-positive(1%-10%)breast cancer as test data to distinguish the status.Then the clinicopathological characteristics,ESR1 mRNA expression level and prognosis of the two groups were compared and analyzed.Results There were significant differences between ER-negative and ER-positive breast cancers in terms of histological type(P=0.01),lymph node metastasis(P=0.02),histological grade(P<0.001),PR(P<0.001),HER2(P<0.001)and Ki-67(P<0.001)expression.The Naive Bayes model was validated on the training cohort,through 5-fold cross-validation,and achieved an outstanding performance of ER status discrimination(AUC=0.91±0.03).The Naive Bayes model was trained on the training cohort and evaluated the 260 patients with ER-low-positive breast cancer.206 patients(79.23%)were classified as ER-negative group,and 54 patients(20.77%)were classified as ER-positive group.By comparing the two groups,the patients in ER-negative group had low levels of ESR1 mRNA expression,little benefit from endocrine therapy and poor prognosis.Conclusion Through the retrospective studies of ER-low-positive cases,the machine learning algorithm can provide a promising perspective for distinguishing the status of ER-low-positive breast cancer.It will assist clinicians in making precision therapy decisions.

关 键 词：乳腺肿瘤 ER弱阳性 机器学习 朴素贝叶斯 

分 类 号：R737.9[医药卫生—肿瘤]