Assessment Of Pet Amyloid Quantification Differences By Varying The Reconstruction Protocol

One goal of PET amyloid imaging is to provide a quantitative estimate of amyloid plaques for staging and monitoring patients with Alzheimer's disease (AD). The standard uptake value ratio (SUVR) is the common measure used and is defined as the PET activity in a target region divided by the PET activity in a reference region. This work focuses on how changes in the reconstruction protocol affect the SUVR in both the target and reference regions. Methods: PET/CT florbetapir data from 15 subjects who had varying levels of cortical amyloid deposits were reconstructed 84 times. The variable space that was spanned included the algorithm (FBP, OSEM and PSF), image voxel size, filter kernel size, TOF and the number of iterations. Results: For a cortical average SUVR referenced to whole cerebellum, deviations were in the range of +/- 4% from the reference OSEM 4i24s 5mm values. The individual target cortical regions that were the most variable were the anterior and posterior cingulate, followed by frontal and temporal pole and inferior frontal regions. For the reference regions, the brainstem was consistently the most variable across reconstruction methods, followed by the pons, while whole cerebellum and white matter were the least variable. PSF introduced less variability than either the smoothing kernel or the differences between FBP and OSEM. The image sharpness provided by > 4 iterations appears to have a potential sensitivity advantage provided that measured regions are extremely consistent anatomically. Conclusions: For same subject data from the same scanner, the same reconstruction method should be applied for within-subject longitudinal studies. For group analysis, the same reconstruction method should be applied across subjects to the extent possible. At a minimum, images should be smoothed to a uniform resolution, and the use of at least modest smoothing will reduce variability. For the reference region, whole cerebellum, white matter, or composites are preferable to brainstem or pons. Use of a large target cortical average region is recommended to reduce longitudinal variability but with the caveat that sensitivity will be reduced.