CT减影影像组学模型中肾透明细胞癌和非透明细胞癌鉴别分析  

Differential analysis of clear cell and non-clear cell renal cell carcinoma in CT subtraction imaging model

作　　者：刘浅浅 姬慧君 陈盈秀 顾梦瑶 陈琪 张濬韬 李万湖 LIU Qianqian;JI Huijun;CHEN Yingxiu;GU MengYao;CHEN Qi;ZHANG Juntao;LI Wanhu(Shandong First Medical University and Shandong Academy of Medical Sciences,Jinan250117,China;Department of Radiology,Shandong Cancer Hospital and Institute,Shandong First Medical University and Shandong Academy of Medical Sciences,Jinan250117,China;Medical Affairs,GE Healthcare PDX,Shanghai 210000,China)

机构地区：[1]山东第一医科大学(山东省医学科学院)研究生院,山东济南250117 [2]山东省肿瘤防治研究院(山东省肿瘤医院)影像科,山东第一医科大学(山东省医学科学院),山东济南250117 [3]通用电气药业(上海)有限公司医学事务部,上海210000

出　　处：《中华肿瘤防治杂志》2023年第8期476-482,共7页Chinese Journal of Cancer Prevention and Treatment

基　　金：山东省肿瘤医院临床研究培育项目(2020-19);北京康盟慈善基金会伦琴影像科研专项(SD-202008-017);中国红十字基金会医学赋能公益专项基金2022年领航菁英临床科研项目(XM_LHJY2022_05_29)。

摘　　要：目的探讨基于肾脏CT皮髓质期-平扫减影的影像组学特征鉴别肾透明细胞癌(ccRCC)和非透明细胞癌(non-ccRCC)的可行性。方法回顾性分析2019-01-01-2021-12-31在山东省肿瘤医院行肾脏CT平扫及增强扫描的114例不同肾细胞癌(RCC)分型患者临床资料,其中男80例,女34例。病理确诊ccRCC 81例,non-ccRCC 33例。采用ITK软件在肾脏CT平扫、皮髓质期以及皮髓质期-平扫减影图像中手动逐层勾画感兴趣区域(ROI),随后提取影像组学特征,采用最大相关最小冗余(mRMR)及最小绝对收缩和选择算子(LASSO)回归进行特征筛选与模型构建,建立平扫、皮髓质期、皮髓质期-平扫减影3组logistic回归模型。受试者工作特征曲线(ROC)评估分类模型性能,DeLong检验比较模型间效能差异。结果一般资料显示,不同RCC亚型患者在性别、肿瘤分布、肿瘤最大直径和年龄差异均无统计学意义,均P>0.05。共提取影像组学特征1218个,进行mRMR后对剩余的30个特征进行LASSO分析。经过10折交叉验证后,肾CT平扫、皮髓质期、皮髓质期-平扫减影图像λ值分别为0.082、0.073和0.046,并分别获得9、6和11个最佳影像组学特征。基于3种图像影像组学评分绘制ROC,CT平扫的影像组学模型中训练队列及测试队列曲线下面积(AUC)分别为0.906(95%CI:0.839~0.972)和0.694(95%CI:0.442~0.947);CT皮髓质期AUC分别为0.882(95%CI:0.803~0.962)和0.852(95%CI:0.721~0.983);CT皮髓质期-平扫减影AUC分别为0.931(95%CI:0.877~0.984)和0.847(95%CI:0.716~0.979)。DeLong检验结果显示,3组间差异均无统计学意义,均P>0.05。结论基于肾脏CT皮髓质期-平扫减影影像组学的预测模型在诊断RCC分型中的价值还需扩大样本继续观察。Objective To test the viability of separating clear cell from non-clear cell renal cell carcinomas by using the radiomics of renal computed tomography(CT)corticomedullary-plain subtraction images.Methods Retrospective analysis was performed on the clinical information of 114patients with various forms of renal cell carcinoma(RCC)who underwent renal CT plain and enhanced scans at Shandong Cancer Hospital and Institute between 2019-01-01and 2021-12-31.Among them,80were males and 34were females.According to the pathological results,there were 33cases of non-clear cell renal carcinoma(non-ccRCC)and 81cases of clear cell renal carcinoma(ccRCC).ITK software was employed in order to accomplish the layer-by-layer manual delineation of renal lesions in CT plain phase,corticomedullary phase,and corticomedullary-plain CT subtraction images.Following the extraction of the radiomics characteristics,feature screening and model construction utilizing the maximum relevance minimum redundancy(mRMR)and least absolute shrinkage and selection operator(LASSO)were performed.Ultimately,logistic regression models for the plain phase,corticomedullary phase,and corticomedullary-plain subtraction were developed.The receiver operating characteristic curve(ROC)was used to assess the performance of the classification models,and DeLong tests were used to analyze the efficiency differences between the models.Results Patients with various RCC subtypes(P>0.05)did not differ from one another with respect to gender,tumor location,maximum tumor diameter,or age.After mRMR,LASSO analysis was performed on the remaining 30features,yielding a total of 1218radiomics features.Following 10-fold cross-validation,λvalues of 0.082,0.073and 0.046were obtained for plain,corticomedullary and corti-corticomedullary-plain subtraction images of renal CT,respectively,and 9,6and 11best radiomics features were obtained,respectively.Three radiomics scores were utilized to generate the ROC.The area under the curve(AUC)values for the training cohort and the test cohort i

关 键 词：影像组学 肾细胞癌 减影 CT 鉴别 

分 类 号：R737.11[医药卫生—肿瘤]