Abstract
2365
Objectives With medium/high energy isotopes, septal penetration in CDR cannot be ignored for accurate SPECT quantification. We propose a novel nearly non-parametric approach to modeling the CDR by estimating PSF template to fit to measured PSFs at different distances.
Methods An I-131 point source was measured at 6 distances from the detector. Three models (Gaussian, Gaussian+Exponential (GauExp), and our Gaussian+Bspline PSF template), which were parameterized with amplitude and width, were fitted to each measurement. For PSF template, symmetric and non-negative constraints were applied. Then, each amplitude and width were fitted over distance with exponential / linear functions that are parameterized with distance. The models were evaluated on SPECT measurements of a ‘hot-sphere’ phantom and patient data with body-contouring. PSF at each distance was generated using fitted functions, and then OSEM reconstructions (6 subsets, 100 iterations) were performed with FFT-based correction for CDR by the 3 different models. A Monte Carlo simulation study with XCAT phantom and SIMIND was also performed for evaluation of the proposed approach.
Results Our proposed method yielded PSF fits that better matched the measured PSFs than the typical Gaussian or GauExp models. The ‘star-shaped’ penetration tails were well fitted with PSF template. In reconstruction, our method achieved better recovery coefficients (RCs) on VOIs (5 - 100 cc spheres) than other methods. In the phantom measurements our method yielded up to 18.6% (8.6%) higher RCs than with the Gaussian (GauExp) method. In XCAT simulation with spherical lesions our method yielded up to 12.9% better RCs than Gaussian method. Lastly, in the patient study, our proposed method increased relative lesion counts (counts in VOI to counts in FOV) by 15.8% (8.6%) compared to the Gaussian (GauExp) method.
Conclusions Our PSF template approach for modeling the CDR substantially improves the quantitative accuracy in I-131 SPECT.
Research Support NIH 2RO1 EB00199