Fast reconstruction of 3D PET data with accurate statistical modeling

Abstract

This paper presents the results of combining high sensitivity 3D PET whole-body acquisition followed by fast 2D iterative reconstruction methods based on accurate statistical models. This combination is made possible by Fourier rebinning (FORE), which accurately converts a 3D data set to a set of 2D sinograms. The combination of volume imaging with statistical reconstruction allows improvement of noise-bias trade-offs when image quality is dominated by measurement statistics. The rebinning of the acquired data into a 2D data set reduces the computation time of the reconstruction. For both penalized weighted least squares (PWLS) and ordered-subset EM (OSEM) reconstruction methods, the usefulness of a realistic model of the expected measurement statistics is shown when the data are pre-corrected for attenuation and random and scattered coincidences, as required for the FORE rebinning algorithm. The results presented are based on 3D simulations of whole body scans that include the major statistical effects of PET acquisition and data correction procedures. As the PWLS method requires knowledge of the variance of the projection data, a simple model for the effect of FORE rebinning on data variance is developed.