Gan Liangyu, Ma Mingming, Liu Yinhua, Liu Qian, Xin Ling, Cheng Yuanjia, Xu Ling, Qin Naishan, Jiang Yuan, Zhang Xiaodong, Wang Xiaoying, Ye Jingming
Breast Disease Center, Peking University First Hospital, Beijing, China.
Department of Radiology, Peking University First Hospital, Beijing, China.
Front Oncol. 2021 Dec 21;11:786346. doi: 10.3389/fonc.2021.786346. eCollection 2021.
To develop a clinical-radiomics model based on radiomics features extracted from MRI and clinicopathologic factors for predicting the axillary pathologic complete response (apCR) in breast cancer (BC) patients with axillary lymph node (ALN) metastases.
The MR images and clinicopathologic data of 248 eligible invasive BC patients at the Peking University First Hospital from January 2013 to December 2020 were included in this study. All patients received neoadjuvant chemotherapy (NAC), and the presence of ALN metastases was confirmed through cytology pre-NAC. The data from January 2013 to December 2018 were randomly divided into the training and validation sets in a ratio of 7:3, and the data from January 2019 to December 2020 served as the independent testing set. The following three types of prediction models were investigated in this study. 1) A clinical model: the model was built by independently predicting clinicopathologic factors through logistic regression. 2) Radiomics models: we used an automatic segmentation model based on deep learning to segment the axillary areas, visible ALNs, and breast tumors on post-NAC dynamic contrast-enhanced MRI. Radiomics features were then extracted from the region of interest (ROI). Radiomics models were built based on different ROIs or their combination. 3) A clinical-radiomics model: it was built by integrating radiomics signature and independent predictive clinical factors by logistic regression. All models were assessed using a receiver operating characteristic curve analysis and by calculating the area under the curve (AUC).
The clinical model yielded AUC values of 0.759, 0.787, and 0.771 in the training, validation, and testing sets, respectively. The radiomics model based on the combination of MRI features of breast tumors and visible ALNs yielded the best AUC values of 0.894, 0.811, and 0.806 in the training, validation, and testing sets, respectively. The clinical-radiomics model yielded AUC values of 0.924, 0.851, and 0.878 in the training, validation, and testing sets, respectively, for predicting apCR.
We developed a clinical-radiomics model by integrating radiomics signature and clinical factors to predict apCR in BC patients with ALN metastases post-NAC. It may help the clinicians to screen out apCR patients to avoid lymph node dissection.
基于从MRI提取的影像组学特征和临床病理因素，开发一种临床影像组学模型，用于预测有腋窝淋巴结（ALN）转移的乳腺癌（BC）患者的腋窝病理完全缓解（apCR）。
本研究纳入了2013年1月至2020年12月在北京大学第一医院的248例符合条件的浸润性BC患者的MR图像和临床病理数据。所有患者均接受了新辅助化疗（NAC），并在NAC前通过细胞学检查确认存在ALN转移。将2013年1月至2018年12月的数据按7:3的比例随机分为训练集和验证集，将2019年1月至2020年12月的数据作为独立测试集。本研究调查了以下三种预测模型。1）临床模型：通过逻辑回归独立预测临床病理因素来构建模型。2）影像组学模型：我们使用基于深度学习的自动分割模型，对NAC后动态对比增强MRI上的腋窝区域、可见的ALN和乳腺肿瘤进行分割。然后从感兴趣区域（ROI）提取影像组学特征。基于不同的ROI或其组合构建影像组学模型。3）临床影像组学模型：通过逻辑回归整合影像组学特征和独立的预测临床因素来构建。所有模型均使用受试者工作特征曲线分析并通过计算曲线下面积（AUC）进行评估。
临床模型在训练集、验证集和测试集中的AUC值分别为0.759、0.787和0.771。基于乳腺肿瘤和可见ALN的MRI特征组合的影像组学模型在训练集、验证集和测试集中分别产生了最佳的AUC值0.894、0.811和0.806。临床影像组学模型在训练集、验证集和测试集中预测apCR的AUC值分别为0.924、0.851和0.878。
我们通过整合影像组学特征和临床因素开发了一种临床影像组学模型，用于预测NAC后有ALN转移的BC患者的apCR。它可能有助于临床医生筛选出apCR患者以避免淋巴结清扫。
