摘要
目的 探讨多模态超声征象结合机器学习对乳腺浸润性导管癌(IDC)中人表皮生长因子受体2(HER2)表达的预测价值.方法 回顾性分析华东师范大学附属芜湖医院2020年3月至2022年12月术后病理证实为乳腺IDC患者160例,所有患者均进行常规超声和自动乳腺全容积扫描(ABVS).依据HER2的表达分为HER2阳性组和HER2阴性组.分析乳腺癌HER2阳性的影响因素;受试者操作特征曲线评估随机森林(RF)、逻辑回归(LR)、高斯朴素贝叶斯(GaussianNB)、K近邻(KNN)和支持向量机分类(SVC)机器学习对乳腺癌HER2阳性的预测价值.结果 两组年龄、最大径、回声模式、形态、纵横比、导管扩张、后方回声比较,差异无统计学意义(P＞0.05);两组边界、微钙化、血流分级、腋窝淋巴结肿大、冠状面特征比较,差异有统计学意义(P＜0.05).微钙化、血流分级、腋窝淋巴结肿大、冠状面特征是乳腺癌HER2阳性表达的危险因素(OR=4.077、2.608、3.093、5.734,P＜0.05).RF、LR、GaussianNB、KNN及SVC预测HER2阳性表达的曲线下面积分别是0.857、0.832、0.833、0.835、0.792.结论 基于多模态超声征象结合机器学习对乳腺IDC中HER2表达具有一定的预测价值,其中RF表现最突出.
Abstract
Objective To explore the predictive value of multimodal ultrasound signs combined with machine learning in human epidermal growth factor receptor 2(HER2)expression in breast invasive ductal carcinoma(IDC).Methods A retrospective analysis was performed on 160 patients with breast IDC confirmed by pathology after surgery in Wuhu Hospital,East China Normal University from March 2020 to December 2022,all pa-tients underwent routine ultrasound and automated breast volume scanner(ABVS).According to the HER2 expression,they were divided into HER2-positive group and HER2-negative group.The influencing factors of HER2 positive breast cancer were analyzed;receiver operating charac-teristic curve were used to evaluate the predictive value of random forest(RF),logistic regression(LR),gaussian naive bayes(GaussianNB),K-near-est neighbor(KNN),and support vector classification(SVC)machine learning for HER2 positive breast cancer.Results There were no significant differences in age,maximum diameter,echo pattern,shape,aspect ratio,catheter dilation,and rear echo between two groups(P＞0.05);there were significant differences in boundary,microcalcification,blood flow grade,axillary lymph node enlargement,and coronal features between two groups(P＜0.05).Microcalcification,blood flow grade,axillary lymph node enlargement,and coronal features were risk factors for HER2 positive expres-sion in breast cancer(OR=4.077,2.608,3.093,5.734,P＜0.05).The area under the curve of RF,LR,GaussianNB,KNN,and SVC predicted HER2 positive expression were 0.857,0.832,0.833,0.835,and 0.792,respectively.Conclusion Multimodal ultrasound signs combined with ma-chine learning have certain predictive value for HER2 expression in breast IDC,in which RF is the most prominent.
基金项目
安徽省重点研究与开发计划项目(202104j07020018)
安徽省高校自然科学研究项目(KJ2020A0614)
NETL
NSTL
万方数据