多参数MRI放射组学诺模图鉴别结节型肾透明细胞癌与乏脂血管平滑肌脂肪瘤的可行性

Feasibility of a Radiomics Nomogram of Multiparametric Magnetic Resonance Imaging to Differentiate Fat-Poor Renal Angiomyolipoma from Nodular Renal Clear Cell Carcinoma

目的探究基于MRI多序列放射组学诺模图鉴别小肾脏肿块(直径≤4 cm)中肾透明细胞癌与乏脂血管平滑肌脂肪瘤的价值。资料与方法回顾性分析2017年7月—2022年12月南京医科大学第一附属医院经病理证实的75例患者共78个肾脏肿块,其中56个肾透明细胞癌、22个乏脂血管平滑肌脂肪瘤,按7∶3分为训练集55个和验证集23个。从T2WI和扩散加权成像序列提取放射组学特征,采用t检验和最小绝对收缩和选择算法进行特征选择,构建放射组学模型,并计算放射组学评分。评估临床特点和MRI主观特征建立临床模型,并结合放射组学评分和临床特征构建放射组学诺模图,评估诺模图的校准、辨别和临床实用性。结果每例患者提取2632个放射组学特征,4个特征用于构建放射组学模型。放射组学模型在训练集[曲线下面积(AUC)=0.979,95%CI 0.937~1.000)]和验证集(AUC=0.833,95%CI 0.626~1.000)中具有良好的辨别能力。放射组学诺模图在训练集(AUC=0.988,95%CI 0.963~1.000)和验证集(AUC=0.867,95%CI 0.698~1.000)中具有良好的校准和辨别能力,较测试集中的临床模型(AUC=0.725,95%CI 0.478~0.972)和放射组学模型(AUC=0.833,95%CI 0.626~1.000)的辨别能力更好。决策曲线分析显示,诺模图的临床实用性优于临床因素模型和放射组学特征。结论基于MRI的放射组学诺模图结合放射组学评分和临床因素,可在术前无创鉴别肾透明细胞癌与乏脂血管平滑肌脂肪瘤。

Purpose To investigate the value of MRI multi-sequence-based radiomic nomogram in identifying clear cell renal cell carcinoma from fat-poor renal angiomyolipoma in small renal masses(≤4 cm).Materials and Methods A retrospective analysis was performed for 78 renal masses in 75 patients with pathologically confirmed cases in the First Affiliated Hospital of Nanjing Medical University from July 2017 to December 2022,including 56 cases of renal clear cell carcinoma and 22 cases of fat-deficient angiomyolipoma,and all participants were divided into a training set(n=55)and a validation set(n=23)in a ratio of 7∶3.Radiomics features were extracted from T2WI and diffusion-weighted imaging sequences,and the t-test and minimum absolute shrinkage and selection algorithm were used for feature selection,the radiomics model was constructed,and the radiomics score was calculated.The clinical characteristics and subjective characteristics of MRI were evaluated to establish a clinical model,and the radiomics nomogram was constructed based on the radiomics score and clinical features,and the calibration,discrimination and clinical practicability of the nomogram were evaluated.Results A total of 2632radiomics features were extracted from each patient,and 4 features were used to construct a radiomics model.The radiomics model had good discrimination ability in the training set[area under the curve(AUC)=0.979,95%CI 0.937-1.000)]and the validation set(AUC=0.833,95%CI 0.626-1.000).The radiomics nomogram had good calibration and discrimination ability in the training set(AUC=0.988,95%CI 0.963-1.000)and validation set(AUC=0.867,95%CI 0.698-1.000),which was better than the clinical model(AUC=0.725,95%CI 0.478-0.972)and radiomics model(AUC=0.833,95%CI 0.626-1.000)in the test set.Decision curve analysis showed that the clinical utility of nomogram was better than that of clinical factor model and radiomics features.Conclusion MRI-based radiomics nomogram combined with radiomics scores and clinical factors can be used to non-invasively distinguish clear cell renal cell carcinoma from alipid-deficient angiomyolipoma before surgery.