Influence of different samplings of line-of-response positions in list-mode PET reconstruction

Abstract

95

Objectives: In current PET scanners, measured LORs contain paired detector IDs but no 3D position information. We modeled the 3D end-point positions of LORs in determination of the system matrix and investigated the impact of using different sampling techniques.

Methods: The GATE Monte Carlo simulator was used to model the Philips GXL PET scanner which uses 4 x 6 x 30 mm GSO(Zr) crystals as its detectors. A large number of back to back photon pair emissions in a uniform cylinder covered the field-of-view (FOV) were generated and tracked. These emission-uniformly sampled LORs with exactly known 3D positions were used to calculate true sensitivity image. The common calculation of the sensitivity image is a uniform LOR sampling scheme. We compared three LOR sampling techniques: 1) fixed position at the detector surface center; 2) random position on the uniform 2D surface; and 3) random 3D position in the detector. The 3D position was sampled according to the accurate GATE results. Point sources in different locations within a water cylinder phantom were generated and used to evaluate different LOR samplings in the list-mode EM reconstruction. The attenuation correction was made in the backprojection operation.

Results: The LOR 3D position sampling can reduce variations of the sensitivity image, matching it well with the true sensitivity. Asymmetric point spread functions (PSFs) and mis-positioning errors of ~5 mm in the FOV edge are observed in the samplings 1) and 2). By the sampling 3), the PSFs are symmetric and the mis-positioning errors are negligible; the spatial resolution (FWHM) is constantly ~4.6 mm along axial direction and is from 3.5 to 10.1 mm in the radial range of 0 to 240 mm.

Conclusions: The 3D sampling can provide a fast alternative to the complete Monte Carlo simulation method in determining sensitivity image.