Optimization of septal spacing in hybrid PET using estimation task performance

Abstract

Hybrid PET imaging, a low cost alternative to dedicated PET, is now commonly used in many clinics. The major factors that limit image quality of hybrid positron emission tomography (PET) systems are a limited count-rate capability, and a low coincidence fraction (i.e., trues to singles ratio). Axial collimation or slats are typically used to reduce single (unpaired) photons from impinging on the detector face. Unfortunately, axial collimation also prevents a number of true coincidences from being detected. In this study, we have used a multi-parameter estimation task to study quantitation performance resulting from projection data obtained using a triple-head system with different levels of axial collimator septal spacing. Specifically, the task is the estimation of amplitude (A) and size (R) of a three-dimensional Gaussian function embedded in a uniform cylindrical background. To evaluate task performance, the bias and variance of maximum-likelihood estimates of the parameters are studied by forming an approximate joint probability density function for the estimates. It was observed that as projection data noise increases; joint PDFs of the estimates become more skewed, with a larger difference between the variance of estimates and their minimum variance bound (Cramer-Rao bound). Estimator performance varied as a function of lesion location, which was expected since coincidence sensitivity for rotating hybrid PET systems with axial collimation is position dependent. In general, results showed that approximately 2.0-cm septal spacing provided projection data with the best overall estimator performance for lesions located within the central 20-cm axial field-of-view.