Impact on Image Noise of Incorporating Detector Blurring Into Image Reconstruction for a Small Animal PET Scanner

Abstract

We study the noise characteristics of an image reconstruction algorithm that incorporates a model of the non-stationary detector blurring (DB) for a mouse-imaging positron emission tomography (PET) scanner. The algorithm uses ordered subsets expectation maximization (OSEM) image reconstruction., which is used to suppress statistical noise. Including the non-stationary detector blurring in the reconstruction process [OSEM(DB)] has been shown to increase contrast in images reconstructed from measured data acquired on the fully-3D MiCES PET scanner developed at the University of Washington. As an extension, this study uses simulation studies with a fully-3D acquisition mode and our proposed FORE+OSEM(DB) reconstruction process to evaluate the volumetric contrast versus noise trade-offs of this approach. Multiple realizations were simulated to estimate the true noise properties of the algorithm. The results show that incorporation of detector blurring FORE+OSEM(DB) into the reconstruction process improves the contrast/noise trade-offs compared to FORE+OSEM in a radially dependent manner. Adding post reconstruction 3D Gaussian smoothing to FORE+OSENT and FORE+OSEM(DB) reduces the contrast versus noise advantages of FORE+OSEM(DB).