Qiu Ya, Zhang Xiang, Wu Zhiyuan, Wu Shiji, Yang Zehong, Wang Dongye, Le Hongbo, Mao Jiaji, Dai Guochao, Tian Xuwei, Zhou Renbing, Huang Jiayi, Hu Lanxin, Shen Jun
Department of Radiology, Sun Yat-Sen Memorial Hospital, Sun Yat-sen University, Guangzhou, China.
Guangdong Provincial Key Laboratory of Epigenetics and Gene Regulation of Malignant Tumors, Sun Yat-sen Memorial Hospital, Guangzhou, China.
Front Oncol. 2022 Feb 28;12:811347. doi: 10.3389/fonc.2022.811347. eCollection 2022.
Overtreatment of axillary lymph node dissection (ALND) may occur in patients with axillary positive sentinel lymph node (SLN) but negative non-SLN (NSLN). Developing a magnetic resonance imaging (MRI)-based radiomics nomogram to predict axillary NSLN metastasis in patients with SLN-positive breast cancer could effectively decrease the probability of overtreatment and optimize a personalized axillary surgical strategy.
This retrospective study included 285 patients with positive SLN breast cancer. Fifty five of them had metastatic NSLNs and 230 had non-metastatic NSLNs. MRI-based radiomic features of primary tumors were extracted and MRI morphologic findings of the primary tumor and axillary lymph nodes were assessed. Four models, namely, a radiomics signature, an MRI-clinical nomogram, and two MRI-clinical-radiomics nomograms were established based on MRI morphologic findings, clinicopathologic characteristics, and MRI-based radiomic features to predict the NSLN status. The optimal predictors in each model were selected using the 5-fold cross-validation (CV) method. Their predictive performances were determined by the receiver operating characteristic (ROC) curves analysis. The area under the curves (AUCs) of different models was compared by the Delong test. Their discrimination capability, calibration curve, and clinical usefulness were also assessed.
The 5-fold CV analysis showed that the AUCs ranged from 0.770 to 0.847 for the radiomics signature, from 0.720 to 0.824 for the MRI-clinical nomogram, from 0.843 to 0.932 for the MRI-clinical-radiomics nomogram. The optimal predictive factors in the radiomics signature, MRI-clinical nomogram, and MRI-clinical-radiomics nomogram were one texture feature of diffusion-weighted imaging (DWI), two clinicopathologic features together with one MRI morphologic finding, and the DWI-based texture feature together with the two clinicopathologic features plus the one MRI morphologic finding, respectively. The MRI-clinical-radiomics nomogram with CA 15-3 included achieved the highest AUC compared with the radiomics signature (0.868 . 0.806, 0.001) and MRI-clinical nomogram (0.868 . 0.761; 0.001). In addition, the MRI-clinical-radiomics nomogram without CA 15-3 showed a higher performance than that of the radiomics signature (AUC, 0.852 . 0.806, = 0.016) and the MRI-clinical nomogram (AUC, 0.852 . 0.761, = 0.007). The MRI-clinical-radiomics nomograms showed good discrimination and good calibration. Decision curve analysis demonstrated that the MRI-clinical-radiomics nomograms were clinically useful.
The MRI-clinical-radiomics nomograms developed in our study showed high predictive performance, which can be used to predict the axillary NSLN status in SLN-positive breast cancer patients before surgery.
腋窝前哨淋巴结（SLN）阳性但非前哨淋巴结（NSLN）阴性的患者可能存在腋窝淋巴结清扫术（ALND）过度治疗的情况。开发基于磁共振成像（MRI）的影像组学列线图来预测SLN阳性乳腺癌患者的腋窝NSLN转移，可有效降低过度治疗的概率并优化个性化腋窝手术策略。
这项回顾性研究纳入了285例SLN阳性乳腺癌患者。其中55例有NSLN转移，230例无NSLN转移。提取原发性肿瘤基于MRI的影像组学特征，并评估原发性肿瘤和腋窝淋巴结的MRI形态学表现。基于MRI形态学表现、临床病理特征和基于MRI的影像组学特征建立了四个模型，即影像组学特征、MRI临床列线图和两个MRI临床影像组学列线图，以预测NSLN状态。使用五折交叉验证（CV）方法选择每个模型中的最佳预测因子。通过受试者工作特征（ROC）曲线分析确定其预测性能。通过德龙检验比较不同模型的曲线下面积（AUC）。还评估了它们的鉴别能力、校准曲线和临床实用性。
五折CV分析显示，影像组学特征的AUC范围为0.770至0.847，MRI临床列线图的AUC范围为0.720至0.824，MRI临床影像组学列线图的AUC范围为0.843至0.932。影像组学特征、MRI临床列线图和MRI临床影像组学列线图中的最佳预测因子分别为扩散加权成像（DWI）的一个纹理特征、两个临床病理特征与一个MRI形态学表现、基于DWI的纹理特征与两个临床病理特征加上一个MRI形态学表现。与影像组学特征（0.868. 0.806，0.001）和MRI临床列线图（0.868. 0.761；0.001）相比，包含CA 15-3的MRI临床影像组学列线图的AUC最高。此外，不包含CA 15-3的MRI临床影像组学列线图的性能高于影像组学特征（AUC，0.852. 0.806， = 0.016）和MRI临床列线图（AUC，0.852. 0.761， = 0.
