动态增强MRI瘤内与瘤周影像组学特征对三阴性乳腺癌的诊断价值研究  

Intranodular and perinodular radiomic features of dynamic contrast-enhanced MRI in diagnosis of triple-negative breast cancer

作　　者：陆欢 葛敏[1] 王世威[1] LU Huan;GE Min;WANG Shiwei(Department of Radiology,the First Affiliated Hospital of Zhejiang Chinese Medical University,Hangzhou 310006,China)

机构地区：[1]浙江中医药大学附属第一医院医学影像科,杭州310006

出　　处：《浙江医学》2021年第15期1647-1651,I0006,共6页Zhejiang Medical Journal

基　　金：浙江省医药卫生科技计划项目(2020KY199);浙江省公益技术应用研究资助项目(LGF21H180003)。

摘　　要：目的探讨乳腺动态增强MRI(DCE-MRI)瘤内与瘤周影像组学特征对三阴性乳腺癌(TNBC)的诊断价值。方法选取2012年12月至2017年5月在浙江中医药大学附属第一医院接受治疗的乳腺癌患者143例。其中TNBC患者23例,非TNBC患者120例,均经手术病理证实,并在术前行乳腺DCE-MRI检查。采用3D Slicer软件在动态增强第二期上勾画病灶,瘤周区域选择瘤体向外扩5 mm进行自动勾画,然后手动调整确认勾画范围。按照7:3的比例随机分为训练集与验证集,使用Analysis Kit(AK)分析软件对训练集提取396个影像组学特征,用mRMR算法及LASSO回归分析对影像组学特征进行降维,用优化的特征集建立模型,并对模型性能进行Hosmer-Lemeshow(H-L)检验。结果经过特征降维,瘤内、瘤周及瘤内加瘤周分别剩余6个影像组学特征被用来建立模型。在验证集中,瘤内、瘤周及瘤内加瘤周的模型ROC曲线AUC分别为0.74、0.71、0.81;灵敏度分别为0.83、0.67、0.83;特异度分别为0.64、0.83、0.83。H-L检验的结果也表明3个模型与实际结果有较好的一致性(P>0.05)。结论乳腺联合DCE-MRI瘤内及瘤周影像组学特征有助于提高TNBC的诊断能力。Objective To investigate the diagnostic value of intranodular and perinodular radiomic features of dynamic contrast-enhanced MRI(DCE-MRI)for triple-negative breast cancer(TNBC).Methods A total of 143 breast cancer patients admitted in the First Affiliated Hospital of Zhejiang Chinese Medical University from December 2012 to May 2017 were enrolled in the study.There were 23 cases with triple-negative breast cancer and 120 cases with non-triple-negative breast cancer as confirmed by surgery and pathology,and all cases underwent preoperative DCE-MRI scan.The segmentation was used by 3D Slicer on the second stage of dynamic enhancement of MRI image.The perinodular region was confirmed by the outline of intranodular outward for 5 mm automatically,and then verified manually.The training cohort and verification cohort were randomly set with a ratio of 7:3.The minimum redundancy maximum relevance(mRMR)feature ranking method and the least absolute shrinkage and selection operator(LASSO)classifier were used to eliminate unnecessary features and the models were generated by logistic regression.The performance of the model for diagnosis of TNBC was evaluated by Hosmer-Lemeshow test.Results The remaining 6 imaging features were used to establish the model.In the validation cohort,the area under ROC curve(AUC)of the intranodular,perinodular and intranodular with perinodular model for diagnosis of TNBC was 0.74,0.71 and 0.81,respectively,and the corresponding sensitivity was 0.83,0.67,0.83,and the specificity was 0.64,0.83,0.83,respectively.The results of Hosmer-Lemeshow test also showed that the models were in good agreement with the actual results(P>0.05).Conclusion The combination of intranodular and perinodular radiomic features on DCE-MRI are conducive to improve the diagnostic efficiency for triple-negative breast cancer.

关 键 词：三阴性乳腺癌 影像组学 磁共振成像 瘤内 瘤周 

分 类 号：R737.9[医药卫生—肿瘤] R445.2[医药卫生—临床医学]