Raniga Parnesh, Bourgeat Pierrick, Fripp Jurgen, Acosta Oscar, Villemagne Victor L, Rowe Christopher, Masters Colin L, Jones Gareth, O'Keefe Graeme, Salvado Olivier, Ourselin Sébastien
CSIRO Preventative Health National Research Flagshift ICTC, The Australian e-Health Research Centre-BioMedIA, Royal Brisbane and Women's Hospital, Herston, QLD, Australia.
Acad Radiol. 2008 Nov;15(11):1376-89. doi: 10.1016/j.acra.2008.07.006.
Radiotracers such as (11)C-PiB have enabled the in vivo imaging of amyloid-beta plaques in the brain, one of the histopathologic hallmarks of Alzheimer's disease (AD). Standardized uptake value ratio (SUVR) has become the most common normalization for (11)C-PiB as it does not require dynamic scans or blood sampling. Normalization is performed by computing the ratio of (11)C-PiB retention in the whole brain to that in cerebellar gray matter. However, SUVR is still conducted manually and is time consuming. An automated normalization algorithm is proposed.
Sixty participants from the Australian Imaging Biomarkers and Lifestyle (AIBL) study were used to test the developed algorithm and compare it against manual SUVR. The cohort consisted of participants likely to have AD (n = 20), those with mild cognitive impairment (MCI; n = 20), and normal controls (NC; n = 20). The participants underwent (11)C-PiB PET scans. A subset (n = 15) also underwent magnetic resonance imaging scans. (11)C-PET scans were segmented using an expectation maximization approach with inhomogeneity correction using three-dimensional cubic B-Splines. A cerebellar region was propagated and constrained by segmentation. Comparisons were made between manual and automated SUVR using regional analysis. Receiver-operating characteristic curves were computed for the task of AD-NC classification. Positron emission tomographic segmentations were also compared to co-registered magnetic resonance images of the same patient.
Significant differences in regional means were observed between manual and automated SUVR. However, these changes were highly correlated (r > 0.8 for most regions). Significant differences (P < .05) in regional variances were also observed for the AD and NC subgroups. Area under the curve was 0.84 and 0.89 for manual and automated SUVR, respectively.
The automated normalization technique results in less within-group variance and better discrimination between AD and NC participants.
诸如（11）C - PiB之类的放射性示踪剂已能够对大脑中的β淀粉样蛋白斑块进行活体成像，这是阿尔茨海默病（AD）的组织病理学特征之一。标准化摄取值比率（SUVR）已成为（11）C - PiB最常用的归一化方法，因为它不需要动态扫描或采血。归一化是通过计算全脑与小脑灰质中（11）C - PiB保留量的比率来进行的。然而，SUVR仍需手动操作且耗时。本文提出了一种自动归一化算法。
来自澳大利亚影像生物标志物与生活方式（AIBL）研究的60名参与者用于测试所开发的算法，并将其与手动SUVR进行比较。该队列包括可能患有AD的参与者（n = 20）、轻度认知障碍（MCI；n = 20）者以及正常对照（NC；n = 20）。参与者接受了（11）C - PiB正电子发射断层扫描（PET）。其中一个子集（n = 1十五）还接受了磁共振成像扫描。（11）C - PET扫描使用期望最大化方法进行分割，并使用三维立方B样条进行不均匀性校正。通过分割传播并约束一个小脑区域。使用区域分析对手动和自动SUVR进行比较。计算用于AD - NC分类任务的受试者操作特征曲线。还将正电子发射断层扫描分割结果与同一患者的共配准磁共振图像进行比较。
手动和自动SUVR之间在区域均值上观察到显著差异。然而，这些变化高度相关（大多数区域r > 0.8）。在AD和NC亚组的区域方差中也观察到显著差异（P < 0.05）。手动和自动SUVR的曲线下面积分别为0.84和0.89。
自动归一化技术导致组内方差更小，并且在AD和NC参与者之间具有更好的区分度。
