摘要
目的:探讨多模态磁共振成像(MRI)影像对乳腺黏液癌(MBC)不同病理组织分级,并对肿瘤标志物Ki-67的表达进行预测研究。方法:参考美国放射学院乳腺影像学报告和数据系统(ACR BI-RADS)对乳腺病变的分级标准,对选取在医院就诊且超声诊断为3级(BI-RADS 3)及以上的120例乳腺粘液癌患者,均使用动态增强磁共振(DCE-MRI)和扩散加权成像(DWI)的多模态MRI影像技术进行检测,记录乳腺病灶MRI传统定量指标中表观弥散系数(ADC)和初始强化率(IER);绘制受试者工作特征(ROC)曲线,分析DCE-MRI和DWI序列ROC曲线下面积(AUC)、灵敏度和特异度;提取DCE-MRI病灶区域5个增强后序列(S1~S5)影像蒙片,收集并分析增强后S1~S5序列纹理特征、统计特征和形态特征的病理报告,察看每例患者的MBC组织学分级情况,并通过免疫组化检测出病理标本相对应的Ki-67阳性细胞的百分比,对提取的S0、S2、S5及ADC的图像不同序列的影像特征进行单变量回归分析和多变量回归分析,利用多分类器模型融合预测Ki-67表达和组织学分级的能力。结果:MBC患者组织样本中Ki-67表达与雌激素受体(PR)、绝经情况和年龄均无统计学意义;多模态MRI组织学分级比较中,Ⅱ级与Ⅲ级MBC患者组织标本中Ki-67表达比较差异有统计学意义(x^2=2.151,P<0.05)。多变量逻辑回归分析预测分级任务中S2序列最佳,AUC、特异度和灵敏度分别为0.78、0.648和0.935。Ki-67预测表达中,DWI选择Fisher Score算法时具有最佳的AUC值(0.783),Ki-67的预测结果AUC、特异度和灵敏度分别为0.783、0.778和0.722。结论:相对于单一参数的磁共振图像数据,多模态MRI影像联合特征可以提高组织分级和Ki-67表达的预测性能。
Objective:To investigate the prediction study of the image of multimodal magnetic resonance imaging(MRI)on grading of differently pathological tissue in mucinous breast carcinoma(MBC)and Ki-67 expression of tumor marker.Methods:120 patients with MBC who admitted to hospital and were confirmed as grade 3(BI-RADS 3)by ultrasound as the grading standard of American College of Radiology Breast Imaging Report and Data System(ACR BI-RADS)for breast lesions were selected.And they were examined by multimodal MRI image technique of dynamic contrast enhancement MRI(DCE-MRI)and diffusion weighted imagine(DWI).And apparent diffusion coefficient(ADC)and initial enhancement rate(IER)of MRI traditionally quantitative indexes of breast lesions were recorded.And the receiver operating characteristic(ROC)curve was drawn,and then the area under curve(AUC)of ROC curve,sensitivity and specificity of DCE-MRI and DWI sequence were analyzed.And the masking films of 5 sequence images(S1-S5)post enhancement of DCE-MRI lesion area were extracted,and the pathological report of textural features,statistical characteristic and morphological characteristic of S1-S5 sequence post enhancement were collected and analyzed.And the situation of MBC histological grading of each patients was observed,and the percentage of Ki-67 positive cell corresponded with pathological specimen was tested by immunohistochemistry.And the image feature of different sequence of extracted S0,S2,S5 and ADC were implemented univariate regression analysis and multivariate regression analysis,and multiple classifiers model was used to integrate and predict the abilities of Ki-67expression and histological grading.Results:There were no significant in the correlation between Ki-67 expression and progesterone receptor(PR),menopause situation and age,respectively.In the comparison of histological grade of multimodal MRI,the difference of Ki-67 expression in tissue specimens between MBC patients with grade II and that with grade III was significant(x^2=2.151,P<0.05).In the predicted grading task of multivariate logistic regression analysis,S2 sequence was the best,and the AUC,specificity and sensitivity of S2 sequence were 0.78,0.648 and0.935,respectively.In the prediction of Ki-67 expression,the AUC value was best when DWI chose Fisher Score algorithm,and the AUC,specificity and sensitivity of prediction results of Ki-67 expression were 0.783,0.778 and 0.722,respectively.Conclusion:Compared with MRI data with single-parameter,the combination of multimodal MRI image features can improve the predictive performance of tissue grading and Ki-67 expression.
作者
陈永毅
刘威
曾旭文
CHEN Yong-yi;LIU Wei;ZENG Xu-wen(Radiology Department,Guangzhou Red Cross Hospital,Guangzhou 510220,China;不详)
出处
《中国医学装备》
2020年第11期14-17,共4页
China Medical Equipment
基金
广州市卫生健康科技项目(20191A011016)“Thiostrepton抑制hedgehog通路诱导三阴型乳腺癌分化的机制研究”。