NEC and local image noise in PET imaging

Noise equivalent counts (NEC) are a widely-used metric of PET scanner performance. The NEC formulation recognizes that scattered and random coincidences (and, sometimes, the randoms correction) add noise to the data, and therefore degrade the signal-to-noise performance of the scanner. NEC has been used to facilitate analysis of many issues in PET imaging. However, there have been a number of confusing results regarding the correlation between NEC and image quality, particularly in comparing 2D and 3D whole body PET studies. This work examines the relationship between the global NEC metric and local image noise, defined here as the variance of single-pixel (or small ROI) values across multiple realizations of a particular set of imaging conditions. An analytic model for SNR performance is derived for filtered backprojection, since pixel signal and variance can be computed directly in that case, and the results of that model are compared to simulation studies using both filtered backprojection and iterative reconstruction. The model indicates that in large objects, a low background (scatter plus randoms) imaging scenario produces better central image SNR than a higher sensitivity but high background scenario with the same global NEC