基于数字化乳腺X线影像组学列线图预测三阴性乳腺癌的多中心研究  

Prediction of Triple-Negative Breast Cancer Based on Digital Mammography Radiomics Nomogram:A Multicenter Study

作　　者：谢玉海[1] 马培旗 韩剑剑 王小雷 胡东 马文俊 魏天贤 杨杨 XIE Yuhai;MA Peiqi;HAN Jianjian;WANG Xiaole;HU Dong;MA Wenjun;WEI Tianxian;YANG Yang(不详;Department of Radiology,Fuyang People′s Hospital,Fuyang 236000,China)

机构地区：[1]太和县人民医院/皖南医学院附属太和医院放射影像科,安徽太和236600 [2]阜阳市人民医院放射影像科,安徽阜阳236000 [3]皖南医学院第一附属医院/弋矶山医院放射影像科,安徽芜湖241000

出　　处：《中国医学影像学杂志》2024年第11期1140-1146,共7页Chinese Journal of Medical Imaging

基　　金：皖南医学院科研项目(JXYY202139)。

摘　　要：目的 探讨基于多中心数字化乳腺X线影像组学列线图预测三阴性乳腺癌(TNBC)的临床价值。资料与方法 回顾性分析2016年11月—2022年3月经病理证实的462例乳腺癌患者的数字化乳腺X线图像,其中243例来自皖南医学院弋矶山医院(机构1)、106例来自阜阳市人民医院(机构2)、113例来自太和县人民医院(机构3)。根据免疫组化结果,按照7∶3将机构1和机构2的乳腺癌患者随机分为训练组244例(TNBC 41例,非TNBC 203例)和验证组105例(TNBC 18例,非TNBC 87例),将机构3的113例乳腺癌患者(TNBC 24例,非TNBC 89例)作为外部验证组。对比分析双乳内外斜位和头尾位图像,选取病变面积较大的数字化乳腺X线图像进行图像分割和影像组学特征提取,经特征筛选后利用逻辑回归构建影像组学模型。由临床病理指标和影像组学评分构建列线图。采用受试者工作特征曲线和决策曲线分析评价模型性能,并比较模型间的预测效能。结果 最终筛选4个与TNBC密切相关的影像组学特征构建影像组学模型,该模型在训练组、验证组、外部测试组预测TNBC的曲线下面积、敏感度和特异度分别为0.868、90.24%和72.91%、0.827、72.22%和75.86%、0.837、70.83%和78.65%。影像组学评分联合组织学分级构建的联合模型在训练组、验证组、外部测试组预测TNBC的曲线下面积、敏感度和特异度分别为0.903、80.49%和86.70%、0.890、77.78%和88.51%、0.870、62.50%和85.39%。联合模型对TNBC的预测效能优于单一影像组学模型,训练组和验证组差异有统计学意义(Z=2.061、2.064,P<0.05)、在外部测试组间差异无统计学意义(Z=1.223,P=0.221)。3组决策曲线分析显示列线图预测三阴性乳腺癌的净收益高于影像组学模型。结论 影像组学模型对TNBC的预测具有较高的诊断效能,结合影像组学评分及组织学分级构建的列线图模型可进一步提高预测效能。Purpose To investigate the clinical value of multi-center digital mammography radiomics nomogram model in predicting triple-negative breast cancer(TNBC).Materials and Methods The digital mammograms of 462 patients with pathologically confirmed breast cancer from November 2016 to March 2022 were retrospectively analyzed,including 243 cases from Yijishan Hospital of Wannan Medical College(institution 1),106 cases from Fuyang People′s Hospital(institution 2)and 113 cases from Taihe People′s Hospital(institution 3).According to the results of immunohistochemistry,a total of 349 breast cancer patients in institution 1 and institution 2 were randomly divided into the training group(244 cases,including 41 TNBC and 203 non-TNBC)and the validation group(105 cases,including 18 TNBC and 87 non-TNBC)according to the ratio of 7∶3,113 breast cancer patients(24 TNBC and 89 non-TNBC)from institution 3 were included in the external validation group.Comparing the mediolateral oblique and cranial cauda digital mammography images,the mammography imaging with larger lesion areas were selected,and the image segmentation and radiomics feature extraction were performed.The radiomics model was constructed by using Logistic regression.The clinicopathological parameters and radiomics scores were used to construct a nomogram.Receiver operating characteristic and decision curve analysis were used to evaluate the model performance.To compare The predictive performance between the models was compared.Results Finally,four radiomics features closely related to TNBC were selected to construct an radiomics model.The area under the curve,sensitivity and specificity of TNBC predicted by the radiomics model in training group,validation group and external test group were 0.868,90.24%and 72.91%,0.827,72.22%and 75.86%,0.837,70.83%and 78.65%,respectively.The area under the curve,sensitivity and specificity of TNBC predicted by the combined model in the training group,validation group and external test group were 0.903,80.49%and 86.70%,0.890,77.78%a

关 键 词：乳腺肿瘤 乳房X线摄影术 影像组学 列线图 病理学 外科 预测 

分 类 号：R737.9[医药卫生—肿瘤] R730.44[医药卫生—临床医学]