Comparison of Partial Volume Correction Techniques for Lesions Near High Activity Regions

Partial volume effect (PVE) is a major factor affecting PET quantification as it leads to miscalculation of uptake in lesions, thereby resulting in inaccurate diagnosis and staging. Many techniques are being used to correct for PVE but there is still need for better and more accurate results. This study aims at investigating the extent of, and correcting spill-over effect from high activity regions to the surrounding lesions. A hot region (bladder) was simulated using the XCAT2 digital phantom and different activities were allocated to simulate uptake increase during a typical FDG PET examination. Spherical lesions with diameter 10 mm and fixed activity were placed at variable distances from the hot region to investigate spill-over effect as a function of activity and distance from the hot region. Validation was done using a NEMA physical phantom consisting of a hot bottle surrounded by lower activity spheres. Analytical simulations were carried out using Software for Tomographic Image Reconstruction (STIR) package and all reconstructions were done using Ordered Subset Expectation Maximization (OSEM), incorporating all corrections within the reconstruction. The reconstructed images were corrected for PVE using three techniques: incorporating point spread function (PSF) in the reconstruction (PVC_PSF), Geometric Transfer Matrix technique (PVC-GTM) and a recently proposed reconstruction- based background correction technique (PVC-BC). Region of Interest (ROI) analysis was carried out using standardized uptake values (SUV), mean and bias as figures of merit. The simulation study revealed a significant spill-over effect from the hot region to the surroundings (which is nearly independent of scatter effect), thereby causing uptake overestimation in lesions within 15 mm around the hot region. The variation in lesion uptake is 43.3% without correction, 42.4% with PVC_PSF, 7.7% with PVC-GTM (mean value) and 1.5% with PVC-BC. This effect was further aggravated with the use of post-filter, as the hot region activity spreads further to the surroundings, thereby increasing uptake and impairing lesion visibility. This was in agreement with the phantom experiment, as the estimated activity around the hot bottle was as much as 20 times higher than the original activity with No-PVC, 15 times with PVC_PSF, but only 3 times with PVC-BC. This effect reduces however as we move further away from the hot region, and as iteration increases. We can therefore conclude that the proposed background correction technique was efficient in obtaining a stable lesion uptake and reduced bias, thereby improving lesion quantification and visibility. However, there is need to further validate the performance of the proposed technique using real data.