Department of Radiology, University of Wisconsin School of Medicine and Public Health, E3/366 Clinical Science Center, 600 Highland Ave, Madison, WI 53792-3252.
Department of Urology, University of Wisconsin School of Medicine and Public Health, Madison, WI.
AJR Am J Roentgenol. 2019 Dec;213(6):1259-1266. doi: 10.2214/AJR.19.21182. Epub 2019 Aug 6.
The purpose of this study was to evaluate the utility of CT texture analysis (CTTA) in differentiating low-attenuation renal cell carcinoma (RCC) from renal cysts on unenhanced CT. Ninety-four patients with low-attenuation RCC on unenhanced CT were compared with a cohort of 192 patients with benign renal cysts. CT characteristics (size and minimum, maximum, and mean attenuation) and CTTA features were recorded using an ROI approximately two-thirds the size of the mass. Masses were subjectively assessed by two expert genitourinary readers and two novice readers using a 5-point Likert scale (1 = definite cyst, 5 = definite renal cell carcinoma). Results of first-order CTTA and subjective evaluation were compared using ROC analysis. The group of 94 patients with low-attenuation RCC included 62 men and 32 women (mean age, 58.0 years). On unenhanced CT, the RCC were larger than 10 mm and of a median size of 50 mm with less than or equal to 20 HU (mean attenuation, 16 ± 4 HU). Of the RCC cohort, 83 were clear cell subtype. The cohort of 192 patients included 134 men and 58 women (mean age, 64.7 years) with benign renal cysts greater than 10 mm and a median size of 27 mm and less than or equal to 20 HU (mean attenuation, 9 ± 6 HU). The mean follow-up time was 6.2 years. Mean entropy in the low-attenuation RCC group (4.1 ± 0.7) was significantly higher than in the cyst group (2.8 ± 1.3, < 0.0001). Entropy showed an ROC AUC of 0.89, with sensitivity of 84% and specificity of 80% at threshold 3.9. The AUC was better than subjective evaluation by novice readers (AUC, 0.77) and comparable to subjective evaluation by two expert readers (AUC, 0.90). A model combining the three best texture features (unfiltered mean gray-level attenuation, coarse entropy, and kurtosis) showed an improved AUC of 0.92. High entropy revealed with CTTA may be used to differentiate low-attenuation RCC from cysts at unenhanced CT; this technique performs as well as expert readers.
本研究旨在评估 CT 纹理分析（CTTA）在区分未增强 CT 上低衰减肾细胞癌（RCC）与肾囊肿中的应用价值。比较了 94 例未增强 CT 上低衰减 RCC 患者与 192 例良性肾囊肿患者的 CT 特征（大小和最小、最大及平均衰减值）和 CTTA 特征，ROI 大小约为肿块的三分之二。使用 5 分 Likert 量表（1=肯定是囊肿，5=肯定是肾细胞癌），由两名泌尿生殖系统专家和两名新手阅读者对肿块进行主观评估。使用 ROC 分析比较了一阶 CTTA 和主观评估的结果。94 例低衰减 RCC 患者中，男 62 例，女 32 例（平均年龄 58.0 岁）。在未增强 CT 上，RCC 直径大于 10mm，中位直径为 50mm，衰减值小于或等于 20HU（平均衰减值为 16±4HU）。在 RCC 组中，83 例为透明细胞亚型。192 例患者中，男 134 例，女 58 例（平均年龄 64.7 岁），良性肾囊肿直径大于 10mm，中位直径为 27mm，衰减值小于或等于 20HU（平均衰减值为 9±6HU）。平均随访时间为 6.2 年。低衰减 RCC 组的平均熵值（4.1±0.7）明显高于囊肿组（2.8±1.3，<0.0001）。熵值的 ROC 曲线下面积（AUC）为 0.89，阈值为 3.9 时，敏感度为 84%，特异度为 80%。AUC 优于新手阅读者的主观评估（AUC 为 0.77），与两位专家阅读者的主观评估相当（AUC 为 0.90）。结合三个最佳纹理特征（未滤波平均灰度衰减值、粗熵和峰度）的模型显示 AUC 有所提高，为 0.92。CTTA 显示的高熵值可能有助于在未增强 CT 上区分低衰减 RCC 与囊肿；该技术的性能与专家阅读者相当。
