瘤内及瘤周CT影像组学联合临床及影像学特征预测透明细胞肾细胞癌侵犯肾被膜  

Intratumoral and peritumoral CT radiomics combined with clinical and imaging features for predicting renal capsule invasion of clear cell renal cell carcinoma

作　　者：张晨阳 贺俊红 王鹏飞 张聪 任金武[2] ZHANG Chenyang;HE Junhong;WANG Pengfei;ZHANG Cong;REN Jinwu(Graduate School of Chengde Medical University,Chengde 067000,China;Department of Medical Imaging,Baoding No.1 Central Hospital,Baoding 071000,China;Department of Medical Imaging,Tianjin TEDA Hospital,Tianjin 300457,China)

机构地区：[1]承德医学院研究生学院,河北承德067000 [2]保定市第一中心医院医学影像科,河北保定071000 [3]天津市泰达医院医学影像科,天津300457

出　　处：《中国医学影像技术》2025年第3期447-451,共5页Chinese Journal of Medical Imaging Technology

摘　　要：目的观察基于瘤内及瘤周ROI的CT影像组学联合临床及影像学特征术前预测透明细胞肾细胞癌(ccRCC)侵犯肾被膜的价值。方法回顾性纳入105例ccRCC患者,根据术后病理分为被膜受侵组(n=70)与未受侵组(n=35),同时按8∶2比例划分训练集(n=84,含受侵组56例、未受侵组28例)与测试集(n=21,含受侵组14例、未受侵组7例)。基于组间差异存在统计学意义的临床及CT影像学特征建立临床-影像学模型。分别于平扫(UP)、皮髓质期(CMP)及实质期(NP)增强CT瘤内及瘤周1~6 mm ROI提取并遴选与被膜受侵相关的影像组学特征,基于CMP瘤内ROI特征于6种机器学习算法中获取最优算法,建立单一瘤内或瘤周模型、同期相瘤内结合瘤周组合模型、同范围不同期相两两组合模型等影像组学模型并遴选最优者,联合临床及影像学特征建立联合模型。绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评估各模型效能,并以DeLong检验进行比较。结果合并高血压、出现临床症状及CMP高强化均为ccRCC侵犯肾被膜的独立预测因素,以之构建临床-影像学模型。支持向量机(SVM)为最优算法,以之构建的单一或组合影像组学模型中,CMP瘤周3 mm模型、CMP瘤内模型、NP瘤周4 mm模型、NP瘤内+瘤周4 mm模型及CMP瘤周3 mm+NP瘤周3 mm模型效能相对较高,其间两两差异均无统计学意义(P均>0.05)。CMP瘤周3 mm模型在测试集中的AUC最高,达0.898,为最优影像组学模型;以之联合临床及影像学特征构建的联合模型在训练集的AUC为0.979,高于临床-影像学模型及最优影像组学模型(P均<0.05),而在测试集的效能(0.918)与后二者相当(P均>0.05)。结论CT瘤周影像组学模型与瘤内模型均可于术前有效预测ccRCC侵及肾被膜;联合临床及影像学特征或可进一步提高诊断效能。Objective To observe the value of intratumoral and peritumoral ROI-based CT radiomics combined with clinical and imaging features for preoperatively predicting renal capsule invasion of clear cell renal cell carcinoma(ccRCC).Methods Totally 105 ccRCC patients were retrospectively collected and divided into invasion group(n=70)and non-invasion group(n=35)according to post operation pathology,also divided into training set(n=84,including 56 cases of invasion group and 28 of non-invasion group)and test set(n=21,including 14 cases of invasion group and 7 of non-invasion group)at a ratio of 8∶2.A clinical-imaging model was constructed based on clinical and CT features being significantly different between groups.Radiomics features related to renal capsule invasion were extracted and selected from intratumoral and of 1—6 mm peritumoral ROI on unenhanced phase(UP),corticomedullary phase(CMP)and nephrographic phase(NP)CT images,respectively.The optimal algorithm was selected among 6 machine learning algorisms based on CMP intratumoral ROI.With the optimal and selected features,single intratumoral or peritumoral models,combined intratumoral and peritumoral models within the same phase and combined pairwise models within the same range across different phases images were established.The best performing radiomics model was chosen and integrated with clinical and imaging features to form a combined model.Receiver operating characteristic(ROC)curves were drawn,the area under the curve(AUC)was calculated to evaluate the efficacy of model for predicting renal capsule invasion of ccRCC,which were compared using DeLong's test.Results Hypertension,presence of clinical symptoms and high enhancement degree on CMP images were all independent predicting factors for renal capsule invasion of ccRCC,which were used to establish clinical-imaging model.Support vector machine(SVM)was the optimal algorithm.CMP peritumoral 3 mm model,CMP intratumoral model,NP peritumoral 4 mm model,NP intratumoral+peritumoral 4 mm model and CMP peritumora

关 键 词：癌 肾细胞 体层摄影术 X线计算机 影像组学 

分 类 号：R737.11[医药卫生—肿瘤] R445.3[医药卫生—临床医学]