Impact of the choice of functional regions in targeted fully 3D SPECT reconstruction

Targeted Fully 3D Monte Carlo (TF3DMC) reconstruction is a tomographic reconstruction approach in Emission Tomography. It is derived from the Fully 3D Monte Carlo (F3DMC) approach in which the system matrix is calculated using Monte Carlo simulations. The originality of the TF3DMC reconstruction is that the volume to be reconstructed is irregularly sampled: a conventional voxel sampling is used in functional regions of interest while only average activity values are estimated in regions that are without any particular interest. In this paper, we study the performance of TF3DMC reconstruction as a function of the definition of these functional regions of interest. This study was performed using SPECT simulations of a phantom filled with Tc99m and including cold and hot rods of various diameters. The system matrix was calculated for 4 different sampling, varying in the delineation of the regions around the rods in which voxel sampling was used. The TF3DMC system matrix was inverted either using MLEM or using a maximum-likelihood algorithm allowing negative reconstruction values (NEG-ML). Rod-to-background activity ratios and signal-to-noise ratios were measured in the reconstructed images. Our results suggest that TF3DMC reconstruction is robust with respect to the delineation of the functional regions of interest: drawing loose regions around the structures of interest is sufficient to accurately estimate the activity in these structures. Using NEG-ML instead of MLEM significantly reduced the biases in activity estimates, leading to biases less than 10% whatever the region. Variance was however increased in the images reconstructed using NEG-ML compared to images reconstructed using MLEM.