Space-variant image-based resolution modeling kernels for enhanced whole-body oncology imaging with the HiRez PET/CT scanner

Objectives Modeling resolution effects in projection space during PET image reconstruction can improve resolution and quantification. This work proposes the use of space-variant resolution modeling for the HiRez PET/CT which captures all resolution degrading effects of both the PET acquisition and reconstruction within image-space kernels.This provides a simpler to implement alternative to projection-space methods.

Methods A grid of 120 printed 18F point sources was scanned at 120 positions to measure the resolution kernels. Every reconstructed point source was fitted by a pair of 2-D Gaussians and model parameters were interpolated and extrapolated for the remaining positions in the FOV. Point sources, phantom data and a [11-C]Survivin human dataset were used to compare images reconstructed using the scanner’s software, a space-invariant resolution model and the proposed space-variant model.

Results At 20cm from the center the space-invariant model improved FWHM resolution radially up to 32%(from 5.5mm to 3.7mm) with the space-variant method improving resolution up to 70%(from 5.5mm to 2mm). In the phantom data the smaller spheres can be better differentiated compared to the spatially invariant model. For the clinical data a better delineation of organ structures is seen.

Conclusions The proposed algorithm shows improvements in the recovered activity with a uniform resolution of 2mm across the FOV for the case of reconstructed point sources. Even though most organs are positioned near the center of the FOV significant changes have been observed at the kidney cortex as well as the liver. The algorithm achieves similar resolution to projection-space resolution modeling without requiring detailed knowledge of the scanner geometry