利用磁共振成像动态增强纹理特征预测不同分子亚型乳腺癌  

Dynamic contrast-enhanced MRI texture analysis for distinguishing different molecular subtypes of breast cancer

作　　者：薛珂 丁莹莹[1] 李振辉[1] 吴建萍[1] 谢瑜[1] 田昌平 杨浩澜 李卓琳[1] XUE Ke;DING Yingying;LI Zhenhui;WU Jianping;XIE Yu;TIAN Changping;YANG Haolan;LI Zhuolin(Department of Radiology,Yuman Cancer Hospital,Kuomning 650118,China;Department of Colorectal Surgery,Yuman Cancer Hospital,Kuomning 650118,China)

机构地区：[1]云南省肿瘤医院放射科,云南昆明650118 [2]云南省肿瘤医院结直肠外科,云南昆明650118

出　　处：《实用放射学杂志》2020年第8期1235-1239,共5页Journal of Practical Radiology

基　　金：云南省科技厅科技计划项目(省基础研究计划-昆医联合专项)(2018FE001-066).

摘　　要：目的利用MRI纹理分析区分不同分子亚型乳腺癌.方法回顾性分析353例经病理证实为乳腺浸润性导管癌患者的临床资料及MRI图像,根据免疫组化或荧光原位杂交结果分为Luminal A型、Luminal B型、人类表皮生长因子受体2(HER-2)过表达型与三阴型乳腺癌(TNBC)4型,利用第三方软件在动态增强强化最明显一期手动勾画病灶,所有层面融合为3D-ROI并提取纹理特征.采用秩和检验和单因素Logistic回归初步筛选患者的临床、纹理特征,再使用多因素Logistic回归挑选独立预测因子并建立预测模型,最后用AUC评估模型的预测效能,并用Hosmer-Lemeshow检验对模型的拟合优良进行检验.结果Luminal A型90例,Luminal B型115例,HER-2过表达型66例,TNBC型82例.通过秩和检验和单因素Logistic回归分析初步筛选出25个特征,再进行多因素Logistic回归分析,分别找出各分子亚型的预测因素并建立预测模型.获得了鉴别Luminal A型与非Luminal A型、Luminal B与非Luminal B型、HER-2过表达型与非HER-2过表达型、TNBC与非TNBC的最佳模型,AUC值分别为0.811、0.604、0.770和0.741,且模型拟合效果均较好.结论基于MRI动态增强纹理特征能够区别不同分子亚型乳腺癌,对Luminal A型的鉴别能力最佳.Objective To distinguish dfferent molecular subtypes of breast cancer by MRI texture features.Methods The elinical data and MRI images of 353 patients with breast invasive ductal carcinoma confirmed by pathology were analyzed retrospectively.According to the results of IHC and FISH,they were divided into four molecular subtypes:Luminal A,Luminal B,HER 2 over expression and triple negative breaust cancer(TNBC).The lesions were manually segmented along the edge of lesion and merged into a three-dimensional region of interest(3D ROI)by using third party software.Texture features were extracted from the 3D ROLRank sun test and univariate Logistic regression were used to analyze the cdinical and texture features.Then rmultivariate Logistic regression was used for slecting independent predictive factors and establishing predicting models.Finally,ROC curves were drawn and AUC were calculated to compare the diagnostic performance of each model.The Hosmer-Lemeshow test was performed to test the goodness of model fines.Results Of all the patients,90 were Luminal A,115 were Luminal B,66 were HER-2 over expressed,and 82 were TNBC.25 features were selected for assing the ability of distinguishing molecular subtypes by rank sum test and univariate Logistic regression.Multivariate Logistic regression analysis was performed to select predictive factors and obtain the best model,results showed that the AUC for classifying Luminal A and non-Lurminal A,Luminal B and non-Luminal B,HER-2 over expression and non-HER 2 over expression,TNBC and non-TNBC subtypes were 0.811,0.604,0.770 and 0.741,respectively.The Hosmer-Le meshorw test showed that these models were all ftted good.Condusion Dynamic contrast-enhanced MRI texture features can distinguish different molecular subtypes of breast cancer,especially identify Luminal A type breast cancer.

关 键 词：乳腺癌 纹理分析 分子亚型 磁共振成像 

分 类 号：R737.9[医药卫生—肿瘤] R445.2[医药卫生—临床医学]