The influence of filtered back-projection and iterative reconstruction on partial volume correction in PET

We assess the influence of the reconstruction algorithms [OS-EM for the iterative one vs. a filtered back-projection in Fourier space (DiFT)] on partial volume correction in PET employing a fully 3D 3-compartment MR based PV-correction algorithm. The gray matter voxels in the PET image -- after removal of the white matter and cerebrospinal fluid contribution -- are corrected voxel-by-voxel using the image resolution.

Phantom measurements and one healthy human brain FDG study were carried out. For the OSEM reconstruction, a combination of iteration steps and subset numbers (It/Sub) was used, whereby in case of no-convergence the image resolution had to be fitted. The RESULTS from the DiFT reconstruction were equivalent to those obtained from the OSEM reconstruction with 10/32 combination for objects with widespread activity concentration. For the sphere phantom, the mean recovery based on the actual values achieved 99.2% +/- 1.8 for all spheres and all reconstruction modes and It/sub combinations (except for 2/8). In case of the Hoffman 3D brain phantom the mean recovery of the cortical regions was 101% +/- 1.2 (the increase based on the uncorrected values: 35.5% +/- 1.5), while the subcortical regions reached a mean recovery of 80% with an increase of 43.9% +/- 2.5. For the human data, an increase of the metabolized values of several cortical regions ranged between 42% and 48% independent from the reconstruction mode.

Our data show that the 3-compartment fully 3-D MR based PV-correction is sensitive to the choice of reconstruction algorithms and to the parameter choice. They indicate that despite improved spatial resolution, the use of the iterative reconstruction algorithm for PV-correction results in similar recovery factors when compared to a correction using DiFT reconstruction, insofar the image resolution values are fitted at the It/Sub combinations.