基于T2WI及动态对比增强MRI的影像组学模型预测肾细胞癌亚型  

Prediction of renal cell carcinoma subtype by T2WI and dynamic contrast-enhanced MRI-based radiomics model

作　　者：李增华 夏春华 胡大涛 黄丹丹 冯倩茹 王亚奇 LI Zeng-hua;XIA Chun-hua;HU Da-tao;HUANG Dan-dan;FENG Qian-ru;WANG Ya-qi(Department of Radiology,The Third Affiliated Hospital of Anhui Medical University,Hefei,Anhui 230000,China)

机构地区：[1]安徽医科大学第三附属医院医学影像科,安徽合肥230000

出　　处：《中国临床研究》2023年第1期34-39,共6页Chinese Journal of Clinical Research

基　　金：合肥市科技局科研合作项目(YW201608080003)。

摘　　要：目的探讨从T2WI及动态增强磁共振图像中获得的影像组学特征在区分肾细胞癌(RCC)三种亚型中的价值。方法回顾性收集安徽医科大学第三附属医院2014年3月至2020年4月经术后病理证实的84例RCC且接受术前磁共振成像(MRI)检查患者的临床影像资料,84例中,透明细胞肾细胞癌(ccRCC)46例、乳头状肾细胞癌(pRCC)20例和嫌色细胞肾细胞癌(cRCC)18例。利用3D-Slicer软件在三个序列(T2WI、EN-T1WI皮质期和EN-T1WI髓质期)上对肿瘤三维全层勾画感兴趣区(ROI),利用Python软件从肿瘤体积中提取影像组学特征。使用组内组间相关分析计算每个特征组内组间相关系数(ICC),选取ICC>0.75的特征作为可重复提取的稳定特征。将肿块随机分为训练集和验证集(约6∶4),使用Kruskal-wallis检验筛选出每个MRI序列鉴别RCC亚型的最佳纹理特征,使用Countif函数对特征子集进行筛选,取三个序列的最佳特征,利用所筛选的基于影像组学的最佳特征分别建立T2WI、EN-T1WI皮质期和EN-T1WI髓质期三个序列的logistic回归模型。报告测试集三种亚型在三个序列的曲线下面积(AUC)、敏感度和特异度。结果三种亚型在三个序列有显著差异的影像组学特征共16个,T2WI、EN-T1WI皮质期、EN-T1WI髓质期序列在区分ccRCC和pRCC时的AUC分别为0.833、0.895和0.885;区分ccRCC和cRCC时的AUC分别为0.822、0.856和0.766;区分pRCC和cRCC时的AUC分别为0.857、0.881和0.857。结论基于MRI中所获得的影像组学资料,T2WI、EN-T1WI皮质期和EN-T1WI髓质期影像组学模型都可以很好的区分ccRCC、pRCC和cRCC,且以EN-T1WI皮质期诊断效能最佳。Objective To investigate the value of radiomics features obtained from T2-weighted imaging(T2WI)and dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)in differentiating the renal cell carcinoma(RCC)subtypes.Methods The clinical and imaging data were retrospectively collected from 84 RCC patients confirmed by postoperative pathology and preoperative MRI from March 2014 to April 2020 in the Third Affiliated Hospital of Anhui Medical University.There were 46 cases of clear cell renal cell carcinoma(ccRCC),20 cases of papillary renal cell carcinoma(pRCC)and 18 cases of chromophobe renal cell carcinoma(cRCC).The three-dimensional full-layer region of interest(ROI)of whole tumor was delineated on three sequences(T2WI,EN-T1WI in cortical phase and EN-T1WI in medullary phase)using 3D Slicer software,and the radiomics features were extracted from the tumor volume using Python software.The correlation analysis was used to calculate intra-group and inter-group correlation coefficient(ICC)for each feature group,and features with ICC values greater than 0.75 were considered reproducible and stable features that could be extracted repeatedly.The lesions were randomly divided into training set and test set according to the proportion of 6∶4,and the best texture features for each MRI sequence to identify RCC subtypes were screened using Kruskal-wallis test.Countif function was used to screen feature subsets for the best features selection of three sequences to establish the logistic regression models of T2WI and EN-T1WI cortical phase and EN-T1WI medullary phase.The AUC,sensitivity and specificity of the three subtypes of the test set in three sequences were calculated and reported.Results There were 16 radiomics features of the three subtypes with significant differences in the three sequences.AUCs of T2WI and EN-T1WI cortical phase and EN-T1WI medullary phase sequence were 0.833,0.895 and 0.885 in distinguishing ccRCC and pRCC,0.822,0.856 and 0.766 for ccRCC and cRCC,and 0.857,0.881 and 0.857 for pRCC and cRCC.Con

关 键 词：肾细胞癌 亚型 磁共振成像 影像组学特征 

分 类 号：R737.11[医药卫生—肿瘤] R445.2[医药卫生—临床医学]