Ma Mingming, Jiang Yuan, Qin Naishan, Zhang Xiaodong, Zhang Yaofeng, Wang Xiangpeng, Wang Xiaoying
Department of Radiology, Peking University First Hospital, Beijing, China.
Beijing Smart Tree Medical Technology Co., Ltd., Beijing, China.
Front Oncol. 2022 Jun 6;12:884599. doi: 10.3389/fonc.2022.884599. eCollection 2022.
To develop a radiomics model based on preoperative dynamic contrast-enhanced MRI (DCE-MRI) to identify sentinel lymph node (SLN) metastasis in breast cancer (BC) patients.
The MRI images and clinicopathological data of 142 female primary BC patients from January 2017 to December 2018 were included in this study. The patients were randomly divided into the training and testing cohorts at a ratio of 7:3. Four types of radiomics models were built: 1) a radiomics model based on the region of interest (ROI) of breast tumor; 2) a radiomics model based on the ROI of intra- and peri-breast tumor; 3) a radiomics model based on the ROI of axillary lymph node (ALN); 4) a radiomics model based on the ROI of ALN and breast tumor. Receiver operating characteristic (ROC) curve analysis and decision curve analysis (DCA) were used to assess the performance of the three radiomics models. The technique for order of preference by similarity to ideal solution (TOPSIS) through decision matrix analysis was used to select the best model.
Models 1, 2, 3, and 4 yielded AUCs of 0.977, 0.999, 0.882, and 1.000 in the training set and 0.699, 0.817, 0.906, and 0.696 in the testing set, respectively, in terms of predicting SLN metastasis. Model 3 had the highest AUC in the testing cohort, and only the difference from Model 1 was statistically significant ( = 0.022). DCA showed that Model 3 yielded a greater net benefit to predict SLN metastasis than the other three models in the testing cohort. The best model analyzed by TOPSIS was Model 3, and the method's names for normalization, dimensionality reduction, feature selection, and classification are mean, principal component analysis (PCA), ANOVA, and support vector machine (SVM), respectively.
ALN radiomics feature extraction on DCE-MRI is a potential method to evaluate SLN status in BC patients.
基于术前动态对比增强磁共振成像（DCE-MRI）开发一种放射组学模型，以识别乳腺癌（BC）患者的前哨淋巴结（SLN）转移情况。
本研究纳入了2017年1月至2018年12月期间142例女性原发性BC患者的MRI图像和临床病理数据。患者以7:3的比例随机分为训练组和测试组。构建了四种类型的放射组学模型：1）基于乳腺肿瘤感兴趣区域（ROI）的放射组学模型；2）基于乳腺肿瘤内部及周边ROI的放射组学模型；3）基于腋窝淋巴结（ALN）ROI的放射组学模型；4）基于ALN和乳腺肿瘤ROI的放射组学模型。采用受试者工作特征（ROC）曲线分析和决策曲线分析（DCA）来评估这三种放射组学模型的性能。通过决策矩阵分析采用逼近理想解排序法（TOPSIS）来选择最佳模型。
在预测SLN转移方面，模型1、2、3和4在训练集中的曲线下面积（AUC）分别为0.977、0.999、0.882和1.000，在测试集中分别为0.699、0.817、0.906和0.696。模型3在测试队列中的AUC最高，且与模型1的差异具有统计学意义（P = 0.022）。DCA显示，在测试队列中，模型3预测SLN转移的净效益高于其他三个模型。通过TOPSIS分析的最佳模型是模型3，该方法用于归一化、降维、特征选择和分类的名称分别为均值、主成分分析（PCA）、方差分析（ANOVA）和支持向量机（SVM）。
DCE-MRI上的ALN放射组学特征提取是评估BC患者SLN状态的一种潜在方法。
