MR影像组学在鉴别腮腺腺淋巴瘤与多形性腺瘤中的价值  

Application of MR-based radiomics signature in differentiating parotid gland adenolymphoma and pleomorphic adenoma

作　　者：田强 王玉瑶 韩宇 南海燕 张曦 崔光彬 TIAN Qiang;WANG Yuyao;HAN Yu;NAN Haiyan;ZHANG Xi;CUI Guangbin(Department of Radiology,Tangdu Hospital,the Air Force Medical University,Xi'an 710038,China;School of Biomedical Engineering,the Air Force Medical University,Xi'an 710032,China)

机构地区：[1]空军军医大学唐都医院放射科,陕西西安710038 [2]空军军医大学军事生物医学工程学系,陕西西安710032

出　　处：《实用放射学杂志》2021年第9期1418-1421,共4页Journal of Practical Radiology

基　　金：国家自然科学基金青年项目(81801655).

摘　　要：目的探讨MR图像的影像组学分析技术在鉴别腮腺腺淋巴瘤(AL)和多形性腺瘤(PA)中的价值.方法回顾性纳入82例经病理证实的腮腺肿瘤患者,其中AL29例,PA53例.所有患者均行MRI扫描,序列包括轴位T_(1)WI、T_(2)WI及T_(2)脂肪抑制(T2FS).基于T2WI图像,逐层勾画全肿瘤区,每个患者可以得到1个三维感兴趣区(ROI).针对ROI提取多种影像组学特征.利用基于支持向量机(SVM)的循环特征消除算法进行特征选择,得到最优特征集,并进行SVM分类器的构建.利用受试者工作特征(ROC)曲线评价模型对鉴别AL和PA的效能.结果在维度为82×3×65的特征矩阵中,通过特征选择算法筛选出前19个特征组成最优特征集,用于分类模型的构建.该模型用于鉴别AL和PA的准确性、敏感性、特异性和曲线下面积(AUC)分别为92.7%、86.2%、96.2%和0.955.结论影像组学方法是一种无创性的术前预测工具,可以挖掘和整合MRI的高维信息,对鉴别AL和PA具有良好的分类准确性.Objective To assess the value of MR-based radiomics signature in differentiating parotid gland adenolymphoma(AL)and pleomorphic adenoma(PA).Methods Eighty-two patients with pathology-proven AL 29 patients and PA 53 patients were retrospectively enrolled.All patients underwent MRI examination,including axial T_(1)WI and T_(2)WI,and T_(2)FS.The whole tumor area was delineated on T_(2)WI,and one region of interest(ROI)was obtained for each patient.Radiomics features were extracted from ROI.With the use of a support vector machine(SVM)with recursive feature elimination,the optimal feature subsets were selected and used to construct predictive model.Receiver operating characteristic(ROC)curve analysis was performed to determine the differentiating efficacy.Results Of the feature matrix with dimension of 82×3×65,19 features were selected to construct SVM model for distinguishing AL from PA.The accuracy,sensitivity,specificity,and area under the curve(AUC)were 92.7%,86.2%,96.2%and 0.955.respectively.Conclusion The presented radiomics approach,a noninvasive preoperative classification tool that can mine and integrate high-dimensional information from MRI,shows favorable predictive accuracy for differentiating the AL and PA.

关 键 词：腮腺 腺淋巴瘤 多形性腺瘤 磁共振成像 影像组学 

分 类 号：R739.8[医药卫生—肿瘤] R445.2[医药卫生—临床医学]