Scintillation event localization in hemi-ellipsoid detector for SPECT, a simulation study using Geant4 Monte-Carlo

Abstract

Abstract Purpose: a high sensitivity Cardiac SPECT system using curved crystals with pinhole collimation was proposed previously (Dey, IEEE Trans. Nucl. Sci. 59 (2012) 334). Using hemi-ellipsoid CsI crystals, in simulations, the system improved sensitivity by a factor of 3 over newer generation SPECT systems (DSPECT or GE Discovery) while keeping the resolution comparable (Bhusal et al., Med. Phys. 46 (2018) 116). In this work, we hypothesize that the high curvature detector results in measurable differences in light distribution from events at different depths in the crystal. We rigorously test this by analyzing the scintillation light using Monte-Carlo (Geant4) and propose a statistical event localization method in a hemi-ellipsoid detector. We evaluate the localization error of the algorithm and the back-projected errors in object space. Methods: to develop this localization capability for the proposed design, we used Geant4 to simulate the propagation of scintillation light in a monolithic hemi-ellipsoidal CsI crystal. A look-up table (LUT) was created to map the points inside the crystal to the expected light pattern on the crystal surface using the Geant4 simulation data. Thirteen zones were considered across the crystal. In each zone, gamma-rays were simulated and the resulting photon intensity on the surface was captured, serving as our experimental interactions. An algorithm based on Poisson statistics was developed to limit the search of the experimental gamma-ray event locations into smaller regions of the LUT. The localization was fine-tuned by comparing the light distribution of the gamma interactions in selected patterns from the LUT points and then recorded. The algorithm-localized gamma-ray events were also individually back-projected to the object mid-plane, (expected mid-plane of the heart), and the error at the plane was recorded as well. Results: the light patterns of adjacent LUT points showed visually discernible differences. Excluding some outliers (up to 2%), the localized errors averaged over all the zones was 0.71 (±0.44) mm with a worse case of 1.36 (± 0.67) mm at the apex. Moreover, when back-projected to the midplane of the region of interest for Cardiac SPECT, the errors were <1 mm due to the high system magnification afforded by the apex and other zones. The average back-projection error at the mid-plane of the object was 0.4 mm±0.22 mm. Conclusion: we modeled gamma-event interactions and scintillation light spread in CsI hemi-ellipsoid detector and developed a robust statistical algorithm that localized scintillation events to within 0.71 (±0.44) mm on the average within a hemi-ellipsoid CsI detector. Moreover, due to the high system magnification afforded by the crystal, the hemi-ellipsoid unit was capable of achieving <1 mm average localization in the object space, assuming perfect pinhole collimator resolution recovery. Thus, we show that this high sensitivity system will be able to deliver images with high resolution for Cardiac SPECT. In the future, the application of this may be extended to Brain SPECT and small animal imaging.