基于CT影像组学鉴别多房囊性肾细胞癌和肾细胞癌囊性变的价值  

Value of differentiating multilocular cystic renal cell carcinoma and necrotic cystic renal cell carcinoma based on CT radiomics

作　　者：冯毅 钟海源 麦镨尹 周泳岑 覃思尚 刘凯洋 谢林润 赵嘉闻 方富翔 黎承杨[1] FENG Yi;ZHONG Haiyuan;MAI Puyin;ZHOU Yongcen;QIN Sishang;LIU Kaiyang;XIE Linrun;ZHAO Jiawen;FANG Fuxiang;LI Chengyang(Department of Urology,First Affiliated Hospital of Guangxi Medical University,Nanning,530021,China;First Clinical Medical College,Guangxi Medical University;Medical College of Oncology,Guangxi Medical University;Second Clinical Medical College,Guangxi Medical University)

机构地区：[1]广西医科大学第一附属医院泌尿外科,南宁530021 [2]广西医科大学第一临床医学院 [3]广西医科大学肿瘤医学院 [4]广西医科大学第二临床医学院

出　　处：《临床泌尿外科杂志》2025年第3期263-269,共7页Journal of Clinical Urology

基　　金：国家自然科学基金地区科学基金项目(No:81660125,82460156);国家自然科学基金青年科学基金项目(No:82400898)。

摘　　要：目的:构建基于CT的影像组学模型,用于区分多房囊性肾细胞癌(multilocular cystic renal cell carcinoma,MCRCC)和肾细胞癌囊性变(necrotic cystic renal cell carcinoma,NCRCC)。方法:通过回顾性分析广西医科大学第一附属医院2009年8月—2023年6月的253例经手术后病理证实为MCRCC或NCRCC的患者的CT影像资料,将其随机划分为训练集177例和验证集76例。对从CT图像中提取到的影像组学特征进行选择,然后使用逻辑回归(logistic regression,LR)和支持向量机(support vector machine,SVM)这2种机器学习算法构建影像组学模型。通过绘制受试者工作特征(receiver operating characteristic,ROC)曲线评估模型鉴别MCRCC和NCRCC的效能,并使用临床决策曲线对模型进行评价。结果:最终筛选出平扫期的11个影像组学特征、皮质期的6个影像组学特征以及平扫+皮质期的11个影像组学特征来构建模型。在模型中,SVM模型表现出最高的曲线下面积(area under the curve,AUC),在训练集中,基于平扫期、皮质期和平扫+皮质期构建的SVM模型的AUC值分别为0.883、0.905和0.954;在验证集中,相应的AUC值分别为0.773、0.902和0.926。临床决策曲线结果显示,平扫+皮质期的SVM模型的临床收益最佳。结论:基于CT的影像组学模型在术前能准确鉴别MCRCC和NCRCC,使用SVM算法构建模型表现最佳,这可能有助于临床的诊断。Objective:To construct CT-based radiomics models for distinguishing between multilocular cystic renal cell carcinoma(MCRCC) and necrotic cystic renal cell carcinoma(NCRCC).Methods:A retrospective analysis was conducted on the CT imaging data of 253 patients with postoperative pathological confirmation of MCRCC or NCRCC at First Affiliated Hospital of Guangxi Medical University from August 2009 to June 2023.They were randomly divided into a training set of 177 cases and a validation set of 76 cases.Then,two machine learning algorithms,logistic regression(LR) and support vector machine(SVM),were used to construct the radiomics model.The effectiveness of the model in distinguishing MCRCC and NCRCC was assessed by drawing receiver operating characteristic(ROC)curves,and the model was evaluated using clinical decision curves.Results:Finally,11 radiomics features were selected from the plain scan phase,6 radiomics features from the cortical phase,and 11 radiomics features from the plain scan+cortical phase to construct the model.Among the models,the SVM model showed the highest area under the curve(AUC),with AUC values of 0.883,0.905 and 0.954 for the SVM model based on plain scan phase,cortical phase and plain scan+cortical phase in the training set,and 0.773,0.902 and 0.926 in the validation set,respectively.The results of the clinical decision curve showed that the SVM model of plain scan+cortical phase had the best clinical benefit.Conclusion:CT-based radiomics models can accurately distinguish between MCRCC and NCRCC before surgery,and the SVM algorithm is the best way to construct the model,which may be helpful for clinical diagnosis.

关 键 词：多房囊性肾细胞癌 肾细胞癌囊性变 影像组学 

分 类 号：R737.11[医药卫生—肿瘤]