From the Department of Radiology (K.A.C., S.W.F.), Geisinger Medical Center, Danville Pennsylvania
Department of Imaging Science and Innovation (Y.H.), Geisinger Medical Center, Danville Pennyslvania. Dr Cauley is currently affiliated with Virtual Radiologic, Eden Prairie, Minnesota.
AJNR Am J Neuroradiol. 2020 May;41(5):809-814. doi: 10.3174/ajnr.A6510. Epub 2020 Apr 23.
Though CT is a highly calibrated imaging modality, head CT is typically interpreted qualitatively. Our aim was to initiate the establishment of a reference quantitative database for clinical head CT.
An automated segmentation algorithm was developed and applied to 354 clinical head CT scans with radiographically normal findings (ages, 18-101 years; 203 women) to measure brain volume, brain parenchymal fraction, brain radiodensity, and brain parenchymal radiomass. Brain parenchymal fraction was modeled using quantile regression analysis.
Brain parenchymal fraction is highly correlated with age ( = 0.908 for men and 0.950 for women), with 11% overall brain volume loss in the adult life span (1%/year from 20 to 50 years and 2%/year after 50 years of age). Third-order polynomial quantile regression curves for brain parenchymal fraction were rationalized and statistically validated. Total brain parenchymal radiodensity shows a decline as a function of age (14.9% for men, 14.7% for women; slopes not significantly different, = .760). Age-related loss of brain radiomass (the product of volume and radiodensity) is approximately 20% for both sexes, significantly greater than the loss of brain volume (< .001).
An automated segmentation algorithm has been developed and applied to clinical head CT images to initiate the development of a reference database for quantitative brain CT imaging. Such a database can be subject to quantile regression analysis to stratify patient brain CT scans by metrics such as brain parenchymal fraction, radiodensity, and radiomass, to aid in the identification of statistical outliers and lend quantitative assessment to image interpretation.
CT 是一种高度校准的成像方式，但头部 CT 通常是定性解读的。我们的目的是为临床头部 CT 建立参考定量数据库。
开发了一种自动分割算法，并将其应用于 354 例影像学正常的临床头部 CT 扫描（年龄 18-101 岁，203 例女性），以测量脑容量、脑实质分数、脑放射性密度和脑实质放射性质量。脑实质分数采用分位数回归分析进行建模。
脑实质分数与年龄高度相关（男性为 0.908，女性为 0.950），成人寿命期内脑总体积损失 11%（20-50 岁每年 1%，50 岁后每年 2%）。脑实质分数的三阶多项式分位数回归曲线是合理的，并经过了统计学验证。总脑实质放射性密度随年龄下降（男性 14.9%，女性 14.7%；斜率无显著差异，=0.760）。男女两性的脑质量随年龄下降约 20%，明显大于脑容量的损失（<0.001）。
开发了一种自动分割算法，并将其应用于临床头部 CT 图像，以启动定量脑 CT 成像参考数据库的开发。这样的数据库可以进行分位数回归分析，根据脑实质分数、放射性密度和放射性质量等指标对患者的脑 CT 扫描进行分层，以帮助识别统计异常值，并对图像解读进行定量评估。
