Carbonell Felix, Zijdenbos Alex P, Hempel Evan, Hajós Mihály, Bedell Barry J
Biospective Inc, 1255 Peel Street, Suite 560, Montreal, QC, H3B 2T9, Canada.
Cognito Therapeutics, 1218 Massachusetts Ave., Suite 200, Cambridge, MA, 02138, USA.
EJNMMI Phys. 2025 Mar 14;12(1):23. doi: 10.1186/s40658-025-00740-9.
Estimation of the spatial resolution in real images is extremely important in several fields, including crystallography, optics, microscopy, and tomography. In human PET imaging, estimating spatial resolution typically involves the acquisition of images from a physical phantom, typically a Hoffman phantom, which poses a logistical burden, especially in large multi-center studies. Indeed, phantom images may not always be readily available, and this method requires constant monitoring of scanner updates or replacements, scanning protocol changes, and image reconstruction guidelines to establish a equivalence with scans acquired from human subjects.
We propose a new computational approach that allows estimation of spatial resolution directly from human subject PET images. The proposed technique is based on the generalization of the logarithmic intensity plots in the 2D Fourier domain to the 3D case. The spatial resolution of the image is obtained through the estimated coefficients of a multiple linear regression problem having the logarithm of the squared norm of the Fourier transform as dependent variable and the squared 3D frequencies as multiple predictors.
The proposed approach was applied to a cohort of subjects consisting of [18F]florbetapir amyloid PET images and matching phantoms from a Phase II clinical trial, and a second cohort including β-amyloid, FDG, and tau PET images from the Alzheimer's Disease Neuroimaging Initiative (ADNI) study. The resulting in-plane and axial resolution estimators varied between 3.5 mm and 8.5 mm for both PET and matching phantom images. They also yielded less than one voxel size across-subjects variability in groups of images sharing the same PET scanner model and reconstruction parameters. For human PET images, we also proved that the spatial resolution estimators showed: (1) a very high reproducibility, as measured by intraclass correlation coefficients (ICC > 0.985), (2) a strong cross-tracer linear correlations, and (3) a high within-subject longitudinal consistency, as measured by the maximum difference value between pairs of visits from the same subject.
Our novel approach does not only eliminate the need for surrogate phantom data, but also provides a general framework that can be applied to a wide range of tracers and other imaging modalities, such as SPECT.
Cognito Therapeutics' OVERTURE clinical trial (NCT03556280, 2021-08-24), https://clinicaltrials.gov/study/NCT03556280 .
在包括晶体学、光学、显微镜学和断层扫描等多个领域中，估计真实图像中的空间分辨率极为重要。在人体正电子发射断层显像（PET）成像中，估计空间分辨率通常需要从物理模型获取图像，通常是霍夫曼模型，这带来了后勤负担，尤其是在大型多中心研究中。实际上，模型图像可能并非随时可得，并且这种方法需要持续监测扫描仪的更新或更换、扫描协议的变化以及图像重建指南，以便与从人体受试者获取的扫描建立等效性。
我们提出了一种新的计算方法，可直接从人体受试者的PET图像估计空间分辨率。所提出的技术基于将二维傅里叶域中的对数强度图推广到三维情况。通过一个多元线性回归问题的估计系数来获得图像的空间分辨率，该问题以傅里叶变换平方范数的对数作为因变量，以三维频率平方作为多个预测变量。
所提出的方法应用于一组受试者，包括来自一项II期临床试验的[18F]氟比他哌淀粉样蛋白PET图像及匹配的模型，以及来自阿尔茨海默病神经影像学倡议（ADNI）研究的包括β淀粉样蛋白、氟代脱氧葡萄糖（FDG）和tau PET图像的第二个队列。对于PET图像和匹配模型图像，所得平面内和轴向分辨率估计值在3.5毫米至8.5毫米之间变化。在共享相同PET扫描仪型号和重建参数的图像组中，它们在受试者间的变化也小于一个体素大小。对于人体PET图像，我们还证明空间分辨率估计值显示出：（1）非常高的可重复性，通过组内相关系数测量（ICC＞0.985）；（2）很强的跨示踪剂线性相关性；以及（3）很高的受试者内纵向一致性，通过同一受试者两次检查之间的最大差值测量。
我们的新方法不仅消除了对替代模型数据的需求，还提供了一个可应用于广泛示踪剂和其他成像模态（如单光子发射计算机断层显像（SPECT））的通用框架。
认知疗法公司的序曲临床试验（NCT03556280，2021年8月24日）,https://clinicaltrials.gov/study/NCT03556280
Zotero 插件
