Air Fraction Correction Optimisation in PET Imaging of Lung Disease

Accurate quantification of radiopharmaceutical uptake from lung PET/CT is challenging due to large variations in fractions of tissue, air, blood and water. Air fraction correction (AFC) uses voxel-wise air fractions, which can be determined from the CT acquired for attenuation correction (AC). However, resolution effects can cause artefacts in either of these corrections. In this work, we hypothesise that the resolution of the CT image used for AC should match that of the intrinsic resolution of the PET scanner but should approximate the reconstructed PET image resolution for AFC. Simulations and reconstructions were performed with the Synergistic Image Reconstruction Framework (SIRF) using phantoms with inhomogeneous attenuation (mu) maps, mimicking the densities observed in lung pathologies. Poisson noise was added to the projection data prior to OSEM reconstruction. AC was performed with a smoothed mu-map, the full-width-half-maximum (FWHM) of the 3D Gaussian kernel was varied (0 - 10 mm). Post-filters were applied to the reconstructed AC images (FWHM: 0 - 8 mm). The simulated mu-map was independently convolved with another set of 3D Gaussian kernels, of varying FWHM (0 - 12 mm), for AFC. The coefficient of variation (CV) in the lung region, designed to be homogeneous post-AFC with optimised kernels, and the mean AFC-standardized uptake value (AFC-SUV) in the regions of simulated pathologies were determined. The spatial resolution of each post-filtered image was determined via a point-source insertion-and-subtraction method on noiseless data. Results showed that the CV was minimised when the kernel applied to the mu-map for AC matched that for the simulated PET scanner and the kernel applied to the mu-map for AFC matched the spatial resolution of the reconstructed PET image. This was observed for all post-reconstruction filters and supports the hypothesis. Initial results from Monte Carlo simulations validate these findings