Abstract
1348
Objectives The goal of this study was to investigate the quantitative accuracy of the closed-form least squares solution (LSS) for SPECT imaging.
Methods Our approach includes three major components: (1) separation of conventionally reconstructed (using MLEM) image to several non-overlapping regions of interest (ROIs) encompassing, for example, organs or tumors; (2) generation of separated systems of equations pertaining to each ROI by calculating fragments of an entire system matrix for voxels belonging to ROI and accurate modeling of cross-contamination between ROIs; (3) direct (non-iterative) solution of the obtained systems: generation of the closed-form weighted penalized LSS (WPLSS) for each ROI. We compared the performance of MLEM with up to 500 iterations (method M1) and WPLSS with (i) Tikhonov- (method M2) and Laplacian-type (method M3) penalties; and (ii) 500 different regularization parameters in recovering accurate absolute activity distribution inside three containers from physical phantom experiments with a hybrid SPECT/CT camera. Corrections for attenuation, scatter, and resolution loss were implemented in all methods.
Results WPLSS and MLEM provided similar accuracies of the recovered total activities for three containers: errors of 4.8-9.4% for M1; 7.2-10.3% for M2, and 5.6-9.0% for M3 were measured. The advantage of WPLSS is in generating more accurate (than MLEM) activity distributions: relative voxel-by-voxel deviation from the true activity distribution was 28.0-31.1% for M1; 17.0-23.5% for M2, and 18.2-26.3% for M3. Visual analysis demonstrated that the images reconstructed by M2-M3 when compared with MLEM-based ones (M1) were less affected by a ringing artifact as well as had more uniform distributions inside containers while more noisy background outside them.
Conclusions Based on the quantitative analysis of images reconstructed from the phantom experiment with three containers, we can conclude that ROI-based WPLSS generates images with more accurate activity distribution than conventional MLEM approach