Implementation and evaluation of a 3D OSEM and median root prior (3D OSEM-MRP) reconstruction algorithm

Abstract

A 3D OSEM and Median Root Prior (3D OSEM-MRP) algorithm has been evaluated for the reconstruction of 3D PET studies. The algorithm was implemented using the software package developed during the EU project PARAPET. Evaluation was performed using experimental phantom data simulating in terms of shape and size PET brain studies. For each phantom, high (/spl sim/200 Mcounts) and low (<50 Mcounts) count statistics 3D PET data were acquired. The performances of the algorithm were evaluated by calculating simple figures of merit (e.g. contrast, coefficient of variation, activity ratio between two regions) based on the use of regions of interest. The performances of the 3D OSEM-MRP were compared with those of a "pure" 3D OSEM and of the PROMIS algorithm, using different reconstruction filters. In all the considered experimental situations, 3D OSEM-MRP shows: 1) to converge to a stable solution, 2) to be quantitatively accurate, 3) to be very effective in noise reduction, particularly for low statistics data, 4) to maintain "good" spatial resolution. Compared with the 3D OSEM and PROMIS algorithms, 3D OSEM-MRP provides better or comparable results depending on the configuration parameters used for the reconstruction of the images.