基于纵向时间影像预测乳腺癌新辅助化疗疗效

Prediction of response to neoadjuvant chemotherapy in breast cancer based on longitudinal images

探讨乳腺癌患者化疗前和化疗早期DCE-MRI影像特征与新辅助化疗疗效的关联,并利用影像特征对疗效进行预测。根据Miller-Payne分级,将61例患者划分为21例化疗有反应和40例化疗无反应;分割提取影像病灶区域特征;在留一法交叉验证下,使用支持向量机模型对疗效进行预测分析。在单变量预测分析中,化疗前和化疗早期影像最佳单特征的预测结果AUC值分别为0.749和0.779;而在进行多变量预测分析时,化疗前和化疗早期影像最优特征子集的预测结果AUC值分别为0.779和0.881。结果表明,化疗早期影像较化疗前影像与疗效关联更为显著,在疗效预测中具有更好的预测性能,可为乳腺癌新辅助化疗提供重要参考。

This study aims to demonstrate the effectiveness of dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)in predicting response to neoadjuvant chemotherapy(NACT).To this end,61 patients who underwent NACT were enrolled with the DCE-MRI acquired before and early NACT,which include 21 responders,and 40 nonresponders,according to the Miller-Payne system.The lesion in DCE-MRI was segmented where radiomic features were extracted.Afterwards,the univariate and multivariate predictive model was conducted using the support vector machine(SVM)under the leave-one-out cross-validation(LOOCV).In the univariate analysis,the AUC for the best performance single feature obtained from preoperative images was 0.749,while was 0.779 for early image.In multivariate analysis,the optimal feature subset generated AUCs of 0.779 and 0.881 for preoperative image and early image,respectively.The results show that the DCE-MRI in the early stage of NACT has a higher correlation with the NACT response than that before NACT,with better predictive performance in the prediction of response.This study has provided promising clinical implications for noninvasively predicting neoadjuvant chemotherapy in breast cancer.