乳腺X线影像组学方法预测乳腺癌腋窝淋巴结转移的价值  

The value of mammography-based radiomics for preoperative prediction of axillary lymph node metastasis in breast carcinoma

作　　者：谭红娜 武明辉 周晶 高飞[2] 海金金 张丹丹 史大鹏 王梅云 Tan Hongna;Wu Minghui;Zhou Jing;Gao Fei;Hai Jinjin;Zhang Dandan;Shi Dapeng;Wang Meiyun(Department of Radiology,Henan Provincial People′s Hospital&People′s Hospital of Zhengzhou University,Imaging Diagnosis of Neurological Diseases and Research Laboratory of Henan Province,Zhengzhou 450003,China;Key Laboratory of Imaging and Intelligent Processing,the People′s Liberation Army Information Engineering University,Zhengzhou 450002,China)

机构地区：[1]河南省人民医院,郑州大学人民医院医学影像科,河南省神经疾病影像诊断与研究重点实验室,450003 [2]中国人民解放军信息工程大学成像与智能处理重点实验室,郑州450002

出　　处：《中华放射学杂志》2020年第9期859-863,共5页Chinese Journal of Radiology

基　　金：国家自然科学基金(81401378);河南省卫生厅科技攻关项目(201602221);中国博士后面上项目(2018M632779)。

摘　　要：目的探讨乳腺X线影像组学方法预测乳腺癌腋窝淋巴结转移的价值。方法回顾性分析2013年6月至2017年7月河南省人民医院病理证实为乳腺癌女性患者的临床及X线资料。共入组214例患者,年龄30~85(53±11)岁,并按照3∶1的比例随机分成训练集(n=153)和验证集(n=61)。根据腋窝淋巴结是否转移,分为腋窝淋巴结阳性组99例,阴性组115例。对获得的双乳内外斜位(MLO)和头尾位(CC)X线图像进行病灶分割和特征提取。应用LASSO回归模型分别从CC、MLO和CC联合MLO图像的高维特征中依次筛选出3、9和7个腋窝淋巴结转移相关的组学特征。根据影像组学特征和临床特征构建预测模型。使用10折交叉验证模型的预测能力。结果腋窝淋巴结阳性组病灶大小大于腋窝淋巴结阴性组,差异有统计学意义(t=2.611,P<0.05)。在验证集中,单独CC、MLO、CC联合MLO图像、临床特征及临床特征联合CC和MLO图像的组学特征预测腋窝淋巴结转移效能的受试者操作特征曲线下面积(AUC)值分别为0.680、0.723、0.740、0.558和0.714,其中,CC联合MLO图像的预测效能最大,AUC值均高于单独CC、MLO图像、CC联合MLO图像预测效能。结论乳腺X线组学特征可术前定量预测乳腺癌腋窝淋巴结转移,但仍需扩大样本量进一步验证。Objective To explore the value of mammography-based radiomics for preoperative prediction of axillary lymph node metastasis in breast carcinoma.Methods The clinical and X-ray data of female patients with pathologically confirmed breast cancer in Henan People′s Hospital from June 2013 to July 2017 were analyzed retrospectively.A total of 214 patients,aged 30-85(53±11)years,were randomly divided into training set(n=153)and verification set(n=61)according to the ratio of 3∶1.According to pathological findings of the axillary lymph node metastasis,99 cases were divided into positive group and 115 cases into negative group.The lesions were segmented and extracted in X-ray images of mediolateral oblique(MLO)and cranial caudal(CC).Three,nine and seven axillary lymph node metastasis related histologic features were selected from the high dimensional features of CC,MLO and CC combined MLO images by lasso regression model.According to the characteristics of imaging and clinical characteristics,the prediction model was constructed.The prediction ability of the model was verified by 10%cross validation.Results The lymph node in positive group was larger than negative groups,the difference was statistically significant(t=2.611,P<0.05).In the validation set,the area under curve(AUC)values of CC,MLO,CC combined with MLO images,clinical features and clinical features combined with CC and MLO images were 0.680,0.723,0.740,0.558 and 0.714,respectively.Among them,CC combined with MLO images had the highest prediction efficiency,and AUC values were higher than CC alone,MLO images and CC combined with MLO images.Conclusions Quantitative radiomics features of breast tumor extracted from digital mammograms are helpful for preoperatively predicting axillary lymph node metastasis.Future larger studies are needed to further evaluate these findings.

关 键 词：乳腺肿瘤 淋巴转移 乳房X线摄影术 

分 类 号：R737.9[医药卫生—肿瘤]