中文摘要：

探讨DCE-MRI影像特征与乳腺癌分子分型之间的关联性。回顾性分析60例术前、化疗前行免疫组化病理检查及DCE-MRI检查的乳腺癌恶性患者。根据免疫组化结果,将患者分为4种分子分型。利用计算机,半自动提取DCE-MRI中包括统计特征、形态特征、纹理特征以及动态增强特征的65维影像特征。利用单变量逻辑回归和多变量逻辑回归方法,评估影像特征和分子分型之间的关联性,并建立多元回归模型,最后研究显著影像特征的分布情况。其中,MRI病灶统计特征与luminal A显著相关,病灶和背景的动态增强特征与luminal B、HER2、basal like显著相关（单变量逻辑回归矫正后P〈0.05）。多变量逻辑回归结果显示,luminal A型、HER2型、basal like型都存在与其显著相关的影像特征,回归方程的P值分别为0.004 73、0.002 77、0.011 7。实验结果说明,DCE-MRI特征可作为潜在的乳腺癌分子分型的影像学标记。

英文摘要：

In this work,we investigated the correlation between DCE-MRI features and molecular subtypes in breast cancer. Sixty cases of malignant breast cancer patients with DCE-MRI examination before chemotherapy were retrospectively analyzed. The molecular subtype was confirmed according to the immunohistochemistry results. Firstly,65-dimensional imaging features including statistical characteristics,morphology,textural and dynamic enhancement were extracted from DCE-MRI with computer semi-automatic methods. Then,the correlations of imaging features and molecular subtype were assessed using statistical analyses,including univariate logistic regression and multivariate logistic regression. At the same time,a multiple regression model was established based on above results. Finally,the distribution of significant image features was analyzed. The results of experiments showed that statistical characteristics of lesions were significantly correlated with luminal A,dynamic enhancement of lesions and background were significantly related to luminal B,HER2 and basal like subtype,in which P values were all lower than 0. 05 using univariate logistic regression-adjusted method.Multi-variable logistic regression analysis showed that imaging features were significantly associated with molecular subtypes with P value equaled to 0. 004 73 for luminal A,0. 002 77 for HER2 and 0. 011 7 for basal like. The results suggested DCE-MR imaging features as important candidate marker to divide breast cancer into molecular subtypes.