Institute of Biomedical Engineering and Instrumentation, Hangzhou Dianzi University, Hangzhou 310018, China.
Zhejiang Cancer Hospital, Hangzhou, 310010, China.
Eur J Radiol. 2017 Sep;94:140-147. doi: 10.1016/j.ejrad.2017.06.019. Epub 2017 Jun 28.
To enhance the accurate prediction of the response to neoadjuvant chemotherapy (NAC) in breast cancer patients by using a quantitative analysis of dynamic enhancement magnetic resonance imaging (DCE-MRI).
A dataset of 57 cancer patients with breast DCE-MR images acquired before NAC was used. Among them, 47 patients were Responders, and 10 patients were non-Responders based on the RECIST criteria. The breast regions were segmented on the MR images, and a total of 158 radiomic features were computed to represent the morphologic, dynamic, and the texture of the tumors as well as the background parenchymal features. The optimal subset of features was selected using evolutionary based Wrapper Subset Evaluator. The classifier was trained and tested using a leave-one-out cross-validation (LOOCV) method to classify Responder and non-Responder cases. The area under a receiver operating characteristic curve (AUC) was computed to assess the classifier performance. An additional independent dataset with 46 patients was also included to validate the results.
The evolutionary algorithm (EA)-based method identified optimal subsets comprising 12 image features that were fit for classification for the main cohort. Following the same feature selection procedure, the independent validation dataset produced 11 image features, 7 of which were identical to those from the main cohort. The classifier based on the features yield a LOOCV AUC of 0.910 and 0.874 for the main and the reproducibility study cohort, respectively. If the optimal features in the main cohort were utilized to test performance on the reproducibility cohort, the classifier generated an AUC of 0.713. While the features developed in the reproducibility cohort were applied to test the main cohort, the classifier achieved an AUC of 0.683. The AUC of the averaged receiver operating characteristic (ROC) curve for the two data cohort was 0.703.
This study demonstrated that quantitative analyses of radiomic features from pretreatment breast DCE-MRI data could be used as valuable image markers that are associated with tumor response to NAC.
通过对动态对比增强磁共振成像（DCE-MRI）进行定量分析，提高乳腺癌患者新辅助化疗（NAC）反应的准确预测。
使用了一组 57 例接受 NAC 前乳腺 DCE-MRI 图像的癌症患者数据。其中，根据 RECIST 标准，47 例患者为应答者，10 例患者为无应答者。在 MR 图像上对乳腺区域进行分割，共计算了 158 个放射组学特征，以表示肿瘤的形态、动态和纹理以及背景实质特征。使用基于进化的包装器子集评估器选择最佳特征子集。使用留一交叉验证（LOOCV）方法对分类器进行训练和测试，以对应答者和无应答者病例进行分类。计算受试者工作特征曲线（ROC）下的面积（AUC）以评估分类器性能。还包括了一组 46 例患者的独立数据集来验证结果。
基于进化算法（EA）的方法确定了包含 12 个适合主队列分类的图像特征的最佳子集。采用相同的特征选择程序，独立验证数据集产生了 11 个图像特征，其中 7 个与主队列的特征相同。基于特征的分类器在主队列和重复性研究队列中的 LOOCV AUC 分别为 0.910 和 0.874。如果在主队列中使用最佳特征来测试重复性队列的性能，分类器生成的 AUC 为 0.713。如果在重复性队列中开发的特征用于测试主队列，则分类器的 AUC 为 0.683。两个数据队列的平均接收器操作特征（ROC）曲线的 AUC 为 0.703。
本研究表明，来自预处理乳腺 DCE-MRI 数据的放射组学特征的定量分析可以作为与 NAC 肿瘤反应相关的有价值的图像标志物。
Zotero 插件