Harmonization of PET imaging biomarkers is a critical step in multi-center studies to enable accurate statistical between-groups comparisons by reducing the bias introduced by inter-site imaging protocols and inter-scanner variability.
With this aim, the use of Hoffman phantoms was a pioneering approach aimed at bringing Hoffman phantom images from different scanner models to a common spatial resolution. Despite the Hoffman-based approach having resulted in a useful tool for large studies, it is logistically challenging and costly to implement in the complex settings of multi-center clinical trials.
Overview of our Hoffman phantom-free approach for estimating spatial resolution in PET images. The logarithmic intensity plots are used for estimating the spatial resolution of PET images. Then, a mathematical formula gives the kernel size to blur the image to the pre-defined common target resolution.
Rather than using Hoffman phantom images as a surrogate, we introduce an approach, called SPITFIRE™, that estimates the spatial resolution directly from the actual human brain PET images.
Our Single PET Image To Find Intrinsic Resolution Estimation (SPITFIRE™) approach is based on the generalization of the classical logarithmic intensity plots in 2D Fourier domain to the 3D case by allowing the simultaneous estimation of the image spatial resolution in both the in-plane and axial directions. Once the spatial resolution is estimated, a simple mathematical formula gives the needed kernel size to blur the PET image to the pre-defined common target resolution.
The performance of the SPITFIRE™ approach has been assessed using two different cohorts of PET images. The first cohort consists of [18F]florbetapir amyloid PET images and matching phantoms from a multi-center clinical trial that used different scanner models in its protocol. The second cohort includes amyloid, tau, and FDG PET images from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study.