1 Department of Internal Medicine, Radiology Division, Ribeirao Preto School of Medicine, University of Sao Paulo, Hospital Clinicas, Avenue Bandeirantes 3900, Campus Monte Alegre, Ribeirao Preto 14049-900, Sao Paulo, Brazil.
2 Department of Surgery and Anatomy, Urology Division, Ribeirao Preto School of Medicine, University of Sao Paulo, Sao Paulo, Brazil.
AJR Am J Roentgenol. 2018 Dec;211(6):1227-1233. doi: 10.2214/AJR.17.19159. Epub 2018 Oct 9.
Our purpose was to assess whether histogram analysis of adrenal lesions from a single measurement of mean attenuation and SD, using a threshold of 10% of negative voxels, can replace voxel counting while maintaining diagnostic accuracy.
In a 4-year period, 325 adrenal lesions were detected on CT examinations of 308 consecutive patients. After exclusions, 91 patients with 108 lesions, including 20 malignant lesions and 88 adenomas (defined by histologic results or follow-up), were enrolled. Two observers retrospectively measured lesion size, mean attenuation value, and SD attenuation value and generated a pixel histogram. The 10th percentile (P10) was obtained from the conventional histogram analysis and was also calculated from the following formula: P10 = mean attenuation - (1.282 × SD). Diagnostic accuracies of the mean attenuation criterion, histogram analysis, and calculated 10th percentile were compared.
The study group was composed of 74 patients with 88 adenomas and 17 patients with 20 malignant lesions, including seven adrenocortical carcinomas and 13 metastases; 93.1% of histograms showed normal distribution. The correlation between histogram analysis and calculated 10th percentile was 0.9827 and 0.9843 for reader 1 and 2 (p < 0.00001 for both). For both readers, sensitivity and specificity of the mean attenuation analysis were 65.9% (95% CI, 55.0-75.7%) and 100.0% (95% CI, 83.2-100%). The sensitivity and specificity of histogram analysis and calculated 10th percentile were the same, 87.5% (95% CI, 78.7-93.6%) and 95.0% (95% CI, 75.1-99.8%), for both readers. The increment increase in sensitivity was significant (p < 0.001), whereas the decrease in specificity was not (p = 0.15).
For most adrenal lesions, the pixel attenuation has a gaussian distribution, allowing estimation of 10th percentile with a single measurement. The accuracy of histogram analysis and calculated 10th percentile outperformed the mean attenuation as a diagnostic criterion for nonfunctioning adenomas.
本研究旨在评估使用 10%负像素阈值的单一测量值（即平均衰减值和标准差）的肾上腺病变直方图分析是否可以替代体素计数，同时保持诊断准确性。
在四年期间，对 308 例连续患者的 CT 检查中检测到 325 个肾上腺病变。排除后，91 例患者的 108 个病变（包括 20 个恶性病变和 88 个腺瘤[通过组织学结果或随访定义]）入组。两名观察者回顾性测量病变大小、平均衰减值和标准差衰减值，并生成像素直方图。从常规直方图分析中获得第 10 百分位数（P10），并通过以下公式计算：P10 = 平均衰减值 -（1.282 × SD）。比较平均衰减标准、直方图分析和计算的第 10 百分位数的诊断准确性。
研究组由 74 例 88 个腺瘤和 17 例 20 个恶性病变患者组成，包括 7 例肾上腺皮质癌和 13 例转移瘤；93.1%的直方图呈正态分布。读者 1 和 2 的直方图分析与计算的第 10 百分位数之间的相关性分别为 0.9827 和 0.9843（两者 p < 0.00001）。对于两位读者，平均衰减分析的敏感性和特异性分别为 65.9%（95%CI，55.0-75.7%）和 100.0%（95%CI，83.2-100.0%）。直方图分析和计算的第 10 百分位数的敏感性和特异性相同，分别为 87.5%（95%CI，78.7-93.6%）和 95.0%（95%CI，75.1-99.8%）。敏感性的增加具有显著意义（p < 0.001），而特异性的降低不具有显著意义（p = 0.15）。
对于大多数肾上腺病变，像素衰减呈正态分布，允许通过单次测量估计第 10 百分位数。直方图分析和计算的第 10 百分位数的准确性优于作为无功能腺瘤诊断标准的平均衰减值。
