We developed the SPITFIRE™ approach for estimating the spatial resolution of PET images using a 3D generalization of logarithmic intensity plots.
Our analysis of actual human brain PET data and accompanying Hoffman phantom images revealed spatial resolution values with very low variability among images coming from the same scanner model and reconstruction parameters.
For all cases, the variability across images aggregated in the same group according to their scanner model was less than the corresponding voxel size.
The SPITFIRE™ estimation also showed excellent longitudinal stability from PET images coming from the same subject at different scanning time points. Additionally, it yielded very strong inter-tracer consistency on those subjects scanned at close time points with different radiotracers.
To the best of our knowledge, this work is the first to provide a clear methodology to estimate the spatial resolution of actual PET images without the need to rely on accompanying Hoffman phantom images and the subsequent tedious and biased image-to-phantom matching process.
Our results show that the proposed approach can be readily used in large-scale clinical trials, thereby considerably decreasing the cost and operational effort of acquiring Hoffman phantoms, especially in multi-center studies.