Reconstruction strategies for the HRRT

Abstract
Various strategies were investigated to reconstruct 3D data from the HRRT, the LSO brain PET installed at the MPI. In this octagonal geometry, the sampling is not complete due to the gaps between flat panel detectors. The standard reconstruction scheme uses a gap-filling step followed by Fourier rebinning (FORE) and attenuation-weighted (AW) OSEM. Scatter correction is performed the single scatter simulation (SSS) approximation. The authors propose the use of 3D weighted OSEM schemes which do not require gap-filling or FORE and which are not sensitive to noise and bias due to those approximations. When list mode data were sorted as true coincidences, the attenuation and normalization OSEM scheme was used. The image from the first iteration was corrected for scatter using the SSS algorithm. The scatter image was used at subsequent iterations to estimate a scatter sinogram. The use of a shifted Poison scheme was precluded since information from singles was missing. When list mode data were sorted as prompt and separate delayed coincidences, an ordinary Poisson scheme was implemented. Images obtained with OSEM 3D were better than those obtained with the standard scheme and were reconstructed at high spatial resolution with 3 iterations and 16 subsets in about 13 hours on a Compaq XP1000 (EV67) workstation.

This publication has 11 references indexed in Scilit: