Resolution Model with Asymmetric and Spatially-Variant PSFs for the S-PET System

A resolution model with asymmetric and spatially-variant point spread functions (asv-PSFs) was designed and implemented in the S-PET prototype system at Shenzhen Bay Laboratory (SZBL). The S-PET system consists of 20 detector modules arranged in a decagonal ring, with each module comprising two layers of crystals of different sizes: a top layer consisting of smaller LYSO crystals sized at $0.86 \times 0.86 \times 5 \mathrm{~mm}^{3}$, arranged in a $30 \times 60$ array, and a bottom layer utilizing BGO crystals sized at $1.29 \times 1.29 \times 7.5 \mathrm{~mm}^{3}$, arranged in a $20 \times 40$ array. We performed image reconstruction using a resolution model that incorporates asv-PSFs in the image space. This was done using convolution operations using the code from Customizable and Advanced Software for Tomographic Reconstruction (CASToR). The asv-PSFs model was compared with the isotropic and spatially-invariant PSF (isi-PSF) model, a built-in model in CASToR. Point source imaging experimental results show that at the axial center of the FOV and a 25 mm radial offset, the radial resolution was improved by 53% from 1.92 mm to 0.90 mm. Two Derenzo-style hot-rod phantoms of different sizes, were used separately in GATE Monte Carlo simulation and experimental measurement to evaluate the effectiveness of the asv-PSFs model. Both simulation and measurement demonstrate the efficacy of the asv-PSFs model in correcting the rod shape distortion caused by the depth of interaction (DOI) effect. Furthermore, the isi-PSF model was unable to distinguish among the 0.75 mm rod sources positioned approximately 20 mm from the center of the FOV, while the asv-PSFs model was able to differentiate among them.