基于增强CT影像组学模型在鉴别良恶性小肾肿瘤的价值

The diagnostic value of contrast-enhanced CT based radiomics in differentiating benign and malignant small renal masses

目的探讨增强CT影像组学模型鉴别良恶性小肾肿瘤(直径≤4 cm)的价值。方法选取经病理证实为恶性小肾肿瘤患者151例,良性小肾肿瘤患者49例。采用分层抽样方法将200例患者按照7∶3比例随机划分为训练集140例及验证集60例,分别于肾皮髓质期、实质期及排泄期逐层手动勾画感兴趣区(ROI),提取影像组学特征。采用最小绝对收缩和选择算法(LASSO)进行降维,建立单独期相模型和联合期相模型共7个模型:1)皮髓质期;2)实质期;3)排泄期;4)皮髓质期-实质期;5)皮髓质期-排泄期;6)实质期-排泄期;7)三期联合模型。通过绘制受试者工作特征曲线(ROC)及曲线下面积(AUC)评估模型的诊断效能。使用De-Long检验比较不同模型间的效能差异。结果三期联合模型的鉴别效能(训练集AUC为0.931;验证集AUC为0.870)高于两两联合模型和单独期相模型,差异有统计学意义(P<0.05);皮髓质期-实质期模型的鉴别效能(训练集AUC为0.913,验证集AUC为0.832)高于皮质期-排泄期模型、实质期-排泄期模型和单独期相模型,差异有统计学意义(P<0.05)。单独期相模型中皮髓质期模型鉴别效能最佳,在训练集和验证集的AUC值分别为0.841和0.760。结论三期联合模型及皮髓质期-实质期模型可进一步提高鉴别诊断良恶性小肾肿瘤的效能。

Objective To investigate the diagnostic value of contrast-enhanced CT based on radiomics in differentiating benign and malignant small renal masses(diameter≤4 cm).Methods By retrospective analysis,151 patients with malignant small renal tumors and 49 patients with benign small renal tumors confirmed by pathology in Jining No.1 People’s Hospital from June 2015 to April 2022 were included.The patients were divided into training set(n=140)and validation set(n=60)according to a ratio of 7∶3 by the stratified random sampling method Regions of interest(ROI)that were manually delineated layer by layer in the renal cortex medullary phase,parenchymal phase and excretory phase to extract imaging features.Least absolute shrinkage and selection operator(LASSO)was used to reduce the dimension.Seven models were established:1)corticomedullary phase;2)nephrogenic phase;3)excretory phase;4)corticomedullary-nephrogenic phase;5)nephrogenic-excretory phase;6)corticomedullary-excretory phase,and 7)triple phases.ROC and De-Long test were used to compare the diagnostic performance among the seven models.The AUC of training and validation sets was denoted as T-AUC and V-AUC,respectively.Results The diagnostic performance of triple phases(T-AUC=0.931,V-AUC=0.870)model was higher than that of pairwise combined model and single phase model,and the differences were statistically significant(all P<0.05).The diagnostic performance of corticomedullary-nephrographic(T-AUC=0.913,V-AUC=0.830)was higher than those of nephrographic-excretory,corticomedullary-excretory models and single phase model,and the differences were statistically significan(t all P<0.05).The corticomedullary model had the best efficiency among all independent-phase models.The AUC values of arterial-phase model in the training and validation sets were 0.841 and 0.760,respectively.Conclusion The triple phases model and corticomedullarynephrogenicphase model of contrast-enhanced CT radiomics can be used to differentiate renal small renal masses.