【摘要】 目的 探讨动态对比增强MRI(dynamic contrast-enhanced MRI, DCE-MRI)图像特征联合定量参数对乳腺癌激素受体、人表皮生长因子受体2 (human epidermal growth factor receptor-2, HER-2)及三阴性乳腺癌（triple-negative breast cancer, TNBC）的预测价值。材料与方法 回顾性分析316例行乳腺MRI检查的乳腺癌患者的临床资料、DCE-MRI图像及定量参数，其中242例为训练组，74例为验证组。根据激素受体表达情况分为激素受体阳性组和阴性组；按照HER-2表达情况分为HER-2阳性组和HER-2阴性组；根据激素受体及HER-2表达状态分为TNBC和非三阴性乳腺癌组（non-triple-negative breast cancer, NTNBC）。比较训练组中不同类型乳腺癌MRI图像特征及定量参数的差异，运用logistic回归联合部分影像征象及定量参数预测激素受体阳性、HER-2阳性与TNBC，构建出相应的列线图，利用验证组进行验证，采用受试者工作特征（receiver operator characteristic, ROC）曲线、校正曲线、决策曲线分析（decision curve analysis, DCA）评估预测模型效能。结果 训练组中病灶大小、形状、毛刺、内部强化特征（internal enhancement characteristics, IEC）、非肿块样强化（non-mass enhancement, NME）、子灶、周围血管增多、腋窝淋巴结肿大、乳头改变等乳腺DCE-MRI图像特征在乳腺癌不同免疫组化结果及分子亚型组间存在显著差异（P均<0.05），定量参数中血管外细胞外间隙容积比（volume fraction of extravascular extra vascular space, Ve）则是激素受体阳性组大于阴性组（P<0.001），而TNBC的Ve小于NTNBC (P<0.001)。毛刺（P<0.001）、IEC(P=0.041)、NME(P=0.006)以及腋窝淋巴结肿大（P=0.029）可以鉴别激素受体阳性和阴性乳腺癌。训练组的曲线下面积（area under the curve, AUC）为0.746(95%CI:0.681～0.811)，敏感度、特异度及准确度分别为82.8%、52.9%及72.3%。验证组的AUC为0.829(95%CI:0.730～0.926)，敏感度、特异度及准确度分别为78.7%、74.1%及77.0%。肿块形状（P=0.050）、毛刺（P=0.016）、NME(P=0.013)及腋窝淋巴结肿大（P<0.001）与HER-2阳性乳腺癌显著相关，结合上述特征构建HER-2阳性乳腺癌预测模型的AUC为0.733(95%CI:0.665～0.800)，敏感度、特异度及准确度分别为55.6%、82.0%及73.1%。验证组的AUC为0.649(95%CI:0.507～0.791)，敏感度、特异度及准确度分别为63.6%、65.4%及64.9%。联合病灶大小（P=0.010）、乳头改变（P=0.016）及Ve(P=0.007)构建的TNBC预测模型，训练组的AUC为0.689(95%CI:0.600～0.779)，敏感性、特异度及准确度分别为80.0%、52.5%及57.0%，验证组的预测模型AUC为0.794(95%CI:0.662～0.927)，敏感度、特异度及准确度分别为86.7%、67.8%及71.6%。训练组及验证组中激素受体阳性、HER-2阳性及TNBC预测模型的校正曲线均表明模型的一致性较高。DCA曲线显示，在较大阈值范围内，使用该模型具有的净收益更高。结论 乳腺癌DCE-MRI的影像特征及定量参数与激素受体及HER-2表达状态相关，具备无创预测激素受体阳性、HER-2阳性及TNBC的潜力。更多还原

【Abstract】 Objective: To explore the predictive value of dynamic contrast-enhanced MRI(DCE-MRI) image features combined with quantitative parameters in breast cancer hormone receptor, human epidermal growth factor receptor-2(HER-2) and triple negative breast cancer(TNBC). Materials and Methods: The clinical data, DCE-MRI images, and quantitative parameters of 316 patients with breast cancer who underwent breast MRI were collected retrospectively, 242 patients in the training group and 74 patients in the validation group. According to the expression of hormone receptor, they were divided into two groups: hormone receptor positive group and negative group. HER-2 positive group and HER-2 negative group were determined by HER-2 expression. TNBC group and non-triple negative breast cancer(NTNBC) group devided according to hormone receptor and HER-2 expression status. In training group, the differences of image features and quantitative parameters among different immunohistochemical results and molecular types of breast cancer were compared. Some imaging features and quantitative parameters selected by logistic regression were used to predict hormone receptor positive, HER-2 positive and TNBC, and then the nomogram models were constructed. The verification group was used for verification. Receiver operator characteristic(ROC) curve, calibration curve and decision curve analysis(DCA) were used to evaluate the performance of the prediction model. Results: In the training group, DCE-MRI image features such as lesion size, shape, spiculated margin, internal enhancement characteristics(IEC), non-mass enhancement(NME), sub-focus, increased peripheral vascularity, axillary lymphadenopathy and nipple change were significantly different among the immunohistochemical results and molecular subtypes of breast cancer(all P<0.05). The quantitative parameter volume fraction of extravascular extra vascular space(Ve) of the hormone receptor positive group was larger than that of hormone receptor negative group(P<0.001), while Ve of the TNBC group was smaller than that of NTNBC group(P<0.001). Spiculated margin(P<0.001), IEC(P=0.041), NME(P=0.006) and axillary lymphadenopathy(P=0.029) can distinguish between hormone receptor positive and negative breast cancer. In training group, combined with the above characteristics, a hormone receptor positive breast cancer prediction model was constructed, and the area under curve(AUC) of prediction model was 0.746(95% CI: 0.681-0.811), sensitivity of 82.8%, specificity of 52.9%, accuracy of 72.3%. In verification group,the AUC of 0.829(95% CI: 0.730-0.926), and the sensitivity, specificity and accuracy were 78.7%, 74.1% and 77.0%, respectively. Mass shape(P=0.050), spiculated margin(P=0.016), NME(P=0.013) and axillary lymphadenopathy(P<0.001) were significantly associated with HER-2 positive breast cancer. In training group, the AUC of HER-2 positive breast cancer prediction model combined with these above characteristics was 0.733(95% CI: 0.665-0.800), sensitivity of 55.6%, specificity of 82.0%, accuracy of 73.1%. In verification group, the AUC of HER-2 positive breast cancer prediction model was 0.649(95% CI: 0.507-0.791), the sensitivity of 63.6%, and the specificity of 64.9%, the accuracy of 64.9%. The TNBC prediction model was combined with lesion size(P=0.010), nipple change(P=0.016) and Ve(P=0.007). In training group, the AUC of TNBC prediction model was 0.689(95% CI: 0.600-0.779), and the sensitivity,specificity and accuracy were 80.0%, 52.5% and 57.0%, respectively. The AUC of prediction model in verification group was 0.794(95% CI: 0.662-0.927), The sensitivity, specificity and accuracy were 86.7%, 67.8% and 71.6%, respectively. The calibration curves of hormone receptor positive, HER-2 positive and TNBC predictive models in training group and verification group showed that the consistency of the models was high. The DCA curve shows that these predictive models could be beneficial among a larger threshold range. Conclusions: Some image features and quantitative parameters derived from DCE-MRI are related to the expression of hormone receptor and HER-2, which has the potential to non-invasively predict hormone receptor positive, HER-2 positive and TNBC.更多还原