浸润性三阴性乳腺癌超声影像组学特征与肿瘤生物学特性的关系研究  

The association between molecular biomarkers and ultrasonographic radiomics features for triple negative invasive breast carcinoma

作　　者：李佳伟 方舟 周瑾 童宇洋 时兆婷 常才 郭翌[2] 余锦华[2] 汪源源[2] Li Jiawei;Fang Zhou;Zhou Jin;Tong Yuyang;Shi Zhaoting;Chang Cai;Guo Yi;Yu Jinhua;Wang Yuanyuan(Department of Medical Ultrasound,Fudan University Shanghai Cancer Center/Department of Oncology,Shanghai Medical College,Fudan University,Shanghai 200032,China;Department of Electronic Engineering,Fudan University,Shanghai 200433,China)

机构地区：[1]复旦大学附属肿瘤医院超声科复旦大学上海医学院肿瘤学系,上海200032 [2]复旦大学电子工程系,上海200433

出　　处：《中华超声影像学杂志》2019年第2期137-143,共7页Chinese Journal of Ultrasonography

基　　金：国家自然科学基金(81371575,81627804,81830058).

摘　　要：目的研究三阴性乳腺癌超声影像组学定量特征与其临床、病理及免疫组化特征的关系。方法收集96例手术病理证实的浸润性三阴性乳腺癌病例,按照患者年龄、肿瘤大小、病理组织学级别、Ki67表达水平及人类表皮生长因子受体2(HER-2)评分分组。同时收集患者的术前超声图像,利用基于相位信息的动态轮廓模型进行边缘分割并提取高通量特征。对每一幅超声图像提取的特征共460个。利用K-svd算法对提取的460维度的特征矩阵进行字典学习,以稀疏表示判断所提取的特征与需要分类的变量之间的相关度。将相关度的权重从大到小排序,按照相关度的权重选取权重最大的特征组合数量输入支持向量机(SVM)分类器以决定选取特征的数量。运用径向基核函数进行分析,预测效能用准确性及ROC曲线下面积(AUC)来表示。结果浸润性三阴性乳腺癌的超声高通量特征与肿块的病理学分级、Ki67表达水平及HER-2表达相关性较大,准确性92.2%~96.9%,AUC98.7%~99.9%。对病理学分级预测分组选取82个特征,权重最大的参数为形态特征中的同二阶中心从矩椭圆离心率;对Ki67表达水平预测分组选取100个特征,权重最大的参数为基于边界纹理特征的环形区的标准偏差;对HER-2表达高低预测分组选取85个特征,权重最大的参数为基于领域灰度差分矩阵(NGTDM)纹理特征的强度。结论浸润性三阴性乳腺癌的超声影像组学定量特征与肿瘤的病理及免疫组化特征具有相关性。超声影像组学特征对预测三阴性乳腺癌的生物学特性具有一定的价值。Objective To evaluate the association between quantitative ultrasonographic features and clinical, pathological and immunohistochemical features of triple negative invasive breast carcinoma(TNBC). Methods With the ethical approval, 96 patients who were pathologically confirmed as TNBC were retrospectively reviewed. All patients were sub-grouped according to age, tumor size, pathological grade, Ki67 expression level and human epidermal growth factor receptor 2 (HER-2) score.Ultrasound images were segmented for the breast carcinoma mass using a phase-based active contour model. The high-throughput radiomics features were extracted based on the two-dimensional sonographic features. There were 460 features extracted from each ultrasound image. A series of computer aided algorithms including K-svd algorithm, sparse representation, support vector machine (SVM) and radial basis function were used to determine the high-throughput sonographic features that were highly correlated to clinical, pathological and immunohistochemical features of TNBC. The performance efficacy was expressed by accuracy and area under curve (AUC) of the ROC curve. Results The high-throughput ultrasonographic features of invasive TNBC could predict its pathological grade, Ki67 level and HER-2 score with the accuracy 92.2%-96.9% and AUC 98.7%-99.9%. There were 82 radiomics features selected for predicting the pathological grade of TNBC, the feature with the maximum weight was the elliptic-normalized eccentricity based on morphological features. There were 100 features selected for predicting the Ki67 expression level, the feature with the maximum weight was the standard deviation of the annular region based on the boundary texture features. There were 85 features selected for the prediction of HER-2 score, the most powerful parameter was the intensity based on NGTDM texture features. Conclusions Quantitative high-throughput ultrasonographic features are correlated with the pathological and immunohistochemical characteristics of invasive TNBC. High-

关 键 词：超声检查 乳腺肿瘤 病理学 影像组学 免疫组化 

分 类 号：R737.9[医药卫生—肿瘤]