超声影像组学标签预测三阴性乳腺癌的价值  

Ultrasound-Based Radiomics in Predicting Triple-Negative Breast Cancer

作　　者：王瑛[1] 陈英格 叶素敏 陈东[2] 吴磊[3] 刘再毅[3] 刘敏 WANG Ying;CHEN Yingge;YE Sumin;CHEN Dong;WU Lei;LIU Zaiyi;LIU Min(不详;Department of Ultrasound,Sun Yat-Sen University Cancer Center,State Key Laboratory of Oncology in South China,Collaborative Innovation Center for Cancer Medicine,Guangzhou 510060,China)

机构地区：[1]广州医科大学附属第一医院超声科,广东广州510120 [2]昆明医科大学第三附属医院,云南省肿瘤医院超声科,云南昆明650118 [3]广东省人民医院放射科,广东广州510080 [4]华南肿瘤学国家重点实验室,肿瘤医学省部共建协同创新中心,中山大学肿瘤防治中心超声科,广东广州510060

出　　处：《中国医学影像学杂志》2021年第1期35-38,46,共5页Chinese Journal of Medical Imaging

基　　金：国家重点研发计划项目(2017YFC1309100)。

摘　　要：目的探讨基于常规超声的影像组学标签在术前诊断三阴性乳腺癌(TNBC)的价值。资料与方法回顾性连续收集230例经手术病理证实的肿块型浸润性乳腺癌患者的临床资料和术前超声图像,按照1∶2随机抽样选取TNBC与非TNBC共102例患者纳入本研究。按超声检查时间顺序,将患者分为训练组66例和验证组36例。通过ImageJ软件手动勾画病灶区域,使用Pyradiomics从每个病灶区域中提取1820个特征,经特征筛选后利用逻辑回归模型构建超声影像组学标签。采用受试者工作特征(ROC)曲线评估超声影像组学诊断TNBC的效能。结果最终选取4个关键影像特征用于构建超声影像组学标签。该标签在训练组和验证组中预测TNBC的ROC曲线下面积分别为0.866(95%CI 0.768~0.964)、0.844(95%CI 0.716~0.972)。校准曲线显示该标签在训练集和验证集中均具有较好的校准度(P=0.765、0.895);决策曲线分析证实该标签能辅助临床决策。结论基于超声的影像组学标签在术前诊断TNBC中具有重要的价值。Purpose To explore the value of a radiomics model based on ultrasound imaging in predicting triple-negative breast cancer(TNBC)before surgery.Materials and Methods A total of 230 patients with invasive breast cancer were retrospectively analyzed,all patients underwent breast ultrasound examination before surgery.A total of 102 patients were randomly selected according to the ratio of 1∶2(TNBC vs.non-TNBC).According to the order of examination time,all patients were divided into 66 cases of training group and 36 cases of validation group.ImageJ software was used to manually delineate the lesion area in the ultrasound image along the tumor boundary.Pyradiomics was used to extract 1820 features from each lesion area,and Logistic regression model was used to construct ultrasound imaging radiomics model.The receiver operating characteristic(ROC)curve was used to evaluate the value of ultrasound imaging radiomics model in predicting TNBC.Results Finally,4 key image features were selected to construct ultrasound imaging radiomics model.The area of under the ROC curve of the model in the training group and the validation group were 0.866(95%CI 0.768-0.964)and 0.844(95%CI 0.716-0.972),respectively.The calibration curve showed that the model had a good calibration in both the training and validation groups;besides,the decision curve analysis confirmed that the model could support clinical decision-making.Conclusion Ultrasound-based imaging radiomics model is of great value in predicting TNBC before surgery.

关 键 词：乳腺肿瘤 受体 雌激素 受体 孕酮 受体 表皮生长因子 超声检查 乳房 影像组学 预测 决策曲线 

分 类 号：R737.9[医药卫生—肿瘤] R730.41[医药卫生—临床医学]