CT影像组学鉴别肾透明细胞癌和非透明细胞癌的价值  

Differential diagnosis of clear cell and non-clear cell renal cell carcinomas based on CT radiomics

作　　者：张建东[1] 孔丹 张辉[1] 冯云[1] 单文莉 段绍峰 郭莉莉[1] ZHANG Jiandong;KONG Dan;ZHANG Hui;FENG Yun;SHAN Wenli;DUAN Shaofeng;GUO Lili(CT Room,the Affiliated Huai'an First Hospital of Nanjing Medical University,Huai'an,J iangsu Province 223300,China;GE Healthcare,Shanghai 210000,China)

机构地区：[1]南京医科大学附属淮安第一医院CT室,江苏淮安223300 [2]GE医疗,上海210000

出　　处：《实用放射学杂志》2020年第12期1985-1988,共4页Journal of Practical Radiology

摘　　要：目的探讨CT影像组学特征鉴别肾透明细胞癌(ccRCC)和肾非透明细胞癌(non-ccRCC)的价值,构建准确性较高的预测模型.方法回顾性分析经病理证实的54例ccRCC和25例non-ccRCC.患者术前均行肾脏多期相CT扫描.使用ITK-SNAP软件人工逐层分割各期相病灶,生成三维ROI.对ROI进行高通量特征采集,并以7︰3的比例随机选择训练组和测试组,应用Spearman相关分析和LASSO降维处理进行特征筛选,构建预测模型,用十折交叉验证方式结合ROC曲线验证各模型的预测效能.结果基于平扫期、皮髓质期、实质期、排泄期及增强3期综合数据构建了5个鉴别ccRCC与non-ccRCC的影像组学模型,5个模型在训练组中的AUC分别为0.93(95%CI:0.85~1.00)、0.98(95%CI:0.94~1.00)、0.93(95%CI:0.84~1.00)、0.92(95%CI:0.84~1.00)和0.98(95%CI:0.93~1.00),模型准确度分别为88%、84%、92%、84%和98%;5个模型在测试组中的AUC分别为0.83(95%CI:0.67~0.99)、0.95(95%CI:0.89~1.00)、0.91(95%CI:0.82~1.00)、0.91(95%CI:0.80~1.00)和0.96(95%CI:0.88~1.00),模型准确度分别为73%、89%、81%、89%和93%.增强3期综合模型的AUC值和准确度最高.结论基于CT影像组学特征构建的5个模型对ccRCC和non-ccRCC术前鉴别诊断均有一定价值,增强3期综合模型预测效能最好.Objective To discuss the value of radiomics model for differentiating clear cell renal cell carcinomas(ecRCC)and non-clear cell renal cell carcinomas(non-ccRCC)based on CT images,and to explore an accurate differential diagnosis prediction model.Methods 54 ccRCC and 25 non-ccRCC patients were enrolled in this study.All patients underwent multi-phase CT scans(NC:non-contrast,CM:corticomedullary,N:nephrographic,E:excretory),then CT images were retrospectively analyzed.Each ROI was segmented manually through ITK-SNAP software.All cases were divided into training and testing cohorts in a ratio of 7:3 randomly.Quantitative radiomics features were extracted from each ROI.The Spearman analysis and LASSO dimension reduction process were used to select valuable features and then differential diagnosis prediction models were constructed.ROC curve was used to estimate the prediction efficiency of each model.Results Quantitative radiomics features were extracted and analyzed based on CT images of four-phase CT scans(NC,CM,N,E)and combination of three phase enhancement.Five radiomics models for differentiating ccRCC and non-ccRCC were constructed based on these features.The AUC of five predictive models in training group were 0.93(95%CI:0.85-1.00),0.98(95%CI:0.94-1.00),0.93(95%CI:0.84-1.00),0.92(95%CI:0.84-1.00,0.98)(95%CI:0.93-1.00)and in testing group were 0.83(95%CI:0.67-0.99),0.95(95%CI:0.89-1.00),0.91(95%CI:0.82-1.00),0.91(95%CI:0.80-1.00),0.96(95%CI:0.88-1.00),respectively.The accuracy in training group were 88%,84%,92%,84%,98%and the accuracy in testing group were 73%,89%,81%,89%,93%,respectively.Combination of three-phase enhancement model had higher AUC value and accuracy.Conclusion The prediction model based on CT radiomics has a good performance on differentiating ccRCC from non-ccRCC.The results can provide a reliable basis for clinical diagnosis and help to differentiate them preoperatively,and combination of three phase enhancement model has higher prediction efficiency.

关 键 词：肾细胞癌 透明细胞型 非透明细胞型 影像组学 计算机体层成像 

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