Estimation of relative PET system sensitivity based on image noise

The aim of this study is to investigate if image noise can be used to estimate the relative system sensitivity under different imaging conditions. Image noise was determined in a uniform cylindrical phantom, a Jaszczak phantom with fillable lesions (0.5-20 ml), with varying lesion to background ratios, ranging from 3-1 to 10-1, and an anthropomorphic torso phantom. In the torso phantom, fillable spheres were added to simulate lesions (5-1 lesion to background ratio). The phantoms were all scanned on a 2-D BGO system and an LSO 3-D system. Image noise was estimated using the bootstrap method. For each data set, 250 sinogram replicates were generated, which were reconstructed using FBP and OSEM (2i8s). The images were filtered to a final image resolution of 7 mm. From the reconstructed image sets, mean and SD images were generated. The relative sensitivities based on scatter corrected count rates from a 20 cm uniform cylinder is 5.4:1 (3D-LSO:2D-BGO). Based on the scan time resulting in equivalent noise levels in the uniform cylinder images, the relative system sensitivity reduced to approximately 4.1:1. A further reduction was seen when comparing the %SD in the spheres in the Jaszczak and the torso phantoms, at 3.1:1 and 2.6:1 at sphere to background ratio of 10-1, and 5:1, respectively. These results indicate that image noise may be a better measure to estimate relative system performance since it includes all factors contributing to the final image quality.