乏脂肪型肾血管平滑肌脂肪瘤与肾透明细胞癌的CT影像组学分析  

The CT radiomics analysis of fat-poor angiomyolipoma and clear cell renal cell carcinoma

作　　者：肖寒 管政 庞佩佩 马燕青 XIAO Han;GUAN Zheng;PANG Peipei;MA Yanqing(Zhejiang Chinese Medical University,Hangzhou 310053,China;Department of Radiology,Haining Central Hospital,Jiacing,Zhejiang Province 314408,China;Department of Radiology,Zhejiang Provincial People's Hospital,Hangzhou 310014,China;GE Healthcare,Hangzhou 310007,China)

机构地区：[1]浙江中医药大学,浙江杭州310053 [2]海宁市中心医院放射科,浙江嘉兴314408 [3]浙江省人民医院放射科,浙江杭州310014 [4]GE医疗生命科学部,浙江杭州310007

出　　处：《实用放射学杂志》2022年第9期1495-1498,共4页Journal of Practical Radiology

基　　金：浙江省卫生健康科技计划项目(2022KY040)。

摘　　要：目的 探讨基于肿瘤及瘤周CT影像组学特征在乏脂肪型肾血管平滑肌脂肪瘤(fp-AML)与肾透明细胞癌(ccRCC)鉴别诊断中的应用价值.方法 回顾性选取经手术病理证实的58例fp-AML与172例ccRCC患者,分析其实质期(NP)CT图像.图像经标准化处理后由2名放射科医生采用ITK软件独立勾画肿瘤及瘤周感兴趣区(ROI).通过A.K软件自动计算肿瘤及瘤周影像组学特征,以7∶ 3的比例随机将数据分成训练集及验证集.经特征选择后,构建3个基于CT影像组学的Logistic模型,包括肿瘤(Log-tumoral)、瘤周(Log-peritumoral)和结合肿瘤及瘤周(Log-combined)的影像组学模型.最后,采用受试者工作特征(ROC)曲线评估鉴别效能.结果 基于CT影像组学的Log-tumoral、Log-peritumoral和Log-combined模型在鉴别fp-AML与ccRCC均表现出较好的诊断性能.其ROC曲线的曲线下面积(AUC)分别为训练集0.884、0.817及0.902,验证集0.862、0.782及0.888.结论 Log-tumoral、Log-peritumoral 和 Log-combined 模型均有助于鉴别 fp-AML 和 ccRCC,Log-combined 模型的鉴别效能较优.Objective To discuss the application value of the CT-based tumoral and peritumoral radiomics features in differentiating fat-poor angiomyolipoma(fp-AML)and clear cell renal cell carcinoma(cRCC).Methods Fifty-eight fp-AML and 172 ccRCC patients confirmed by surgery and pathology were retrospectively enrolled,and the nephrographic phase(NP)CT images were analyzed.After standardizing images,two radiologists independently delineated the tumoral and peritumoral region of interest(ROI)using ITK software.The radiomics features were automatically calculated in A.K software and randomly divided into the training set and testing set at a ratio of 7:3.After feature selection,three CT-based radiomics Logistic models were constructed,including Log-tumoral,Log-peritumoral,and Log-combined models.Finally,the receiver operating characteristic(ROC)curves were used to evaluate the distinguished efficacy.Results Log-tumoral,Log-peritumoral,and Log-combined models based on CT radiomics showed good performance in distinguishing fp-AML from ccRCC.And the area under the curve(AUC)of the three models were O.884,0.817 and 0.902 in the training set,and 0.862,0.782 and 0.888 in the testing set,respectively.Conclusion Log-tumoral,Log-peritumoral,and Log-combined models can help differentiating fp-AML from ccRCC.The Log-combined model show better performance in differentiating two diseases.

关 键 词：影像组学 肾血管平滑肌脂肪瘤 肾透明细胞癌 计算机体层成像 

分 类 号：R445[医药卫生—影像医学与核医学] R737.11[医药卫生—诊断学] R814.42[医药卫生—临床医学]