基于CT影像组学列线图鉴别≤4 cm肾透明细胞癌与乏脂肪血管平滑肌脂肪瘤

A CT-Based Radiomics Nomogram for Differentiation of ≤4 cm Clear Cell Renal Cell Carcinoma and Angiomyolipoma without Visible Fat

摘要: 目的：基于CT的影像组学组学特征联合CT图像特征建立影像组学列线图，验证其在≤4 cm肾透明细胞癌(ccRCC)与肾乏脂肪血管平滑肌脂肪瘤(AML.wovf)之间的鉴别效能。方法：回顾性分析149例肾透明细胞癌(n = 102)和肾乏脂肪血管平滑肌脂肪瘤(n = 47)患者。从皮髓期、实质期及排泄期CT图像中提取影像组学特征，并计算影像组学评分(Radscore)。评估CT特征及一般临床因素建立图像特征模型。结合Radscore和独立的图像特征，绘制影像组学列线图，并从拟合度、鉴别效能和临床实用性方面对列线图进行评估。结果：影像组学列线图(曲线下面积(AUC)，0.979；95%置信区间(CI)，0.962~0.996)与影像组学特征(AUC, 0.972; 95%CI, 0.949~0.994)及图像特征模型(AUC, 0.933; 95%CI, 0.897~0.970)相比，显示出更好的鉴别效能。决策曲线分析显示列线图的临床实用性优于影像组学特征和图像特征模型。结论：基于CT的影像组学列线图是一种无创的术前预测工具，在鉴别AML.wovf和ccRCC显示出良好的预测效果，这可能有助于临床医生制定精确的治疗方案。

Abstract: Objective: CT based radiomics features combined with CT image features to establish radi-omicsnomogram and verify its discriminant effectiveness between ≤4 cm clear cell renal cell carci-noma and renal angiomyolipoma without visible fat (AML.wovf). Methods: 149 patients with ccRCC (n = 102) and AML.wovf (n = 47) were retrospectively analyzed. The imaging group characteristics were extracted from the CT images of the corticomedullary, parenchymal and excretory phases, and the radiomic score (Radscore) was calculated. Evaluating CT features and general clinical factors to establish image feature model. Combined with Radscore and independent image features, the ra-diomics nomogram was drawn, and the nomogram was evaluated from the aspects of fitting, dis-crimination efficiency and clinical practicability. Results: Radiomicsnomogram (area under curve (AUC), 0.979; 95% confidence interval (CI), 0.962~0.996) showed better discrimination efficiency compared with the radiomics features (AUC, 0.972; 95%CI, 0.949~0.994) and image features (AUC, 0.933; 95%CI, 0.897~0.970). The decision curve analysis showed that the clinical practicability of nomogram was better than that of radiomics features and CT image features model. Conclusion: CT based radiomics nomogram is a non-invasive preoperative prediction tool, which shows good pre-diction effect in distinguishing AML.wovf and ccRCC, which may help clinicians to formulate accurate treatment plans.