RESampling between Projection SpACEs (RESPACE) using Bayse’ Theorem

In order to simplify the Point-Spread-Function (PSF) reconstruction framework, resolution modelling can be decoupled from the iterative reconstruction process by an additional data resampling step previous to reconstruction. We call the proposed algorithm RESampling between Projection spACEs (RESPACE). In this abstract, RESPACE is applied to resample the simulated projection data to 2D Generic Cylinder Model (GCM) projection data structure, which will be used for reconstruction afterwards. Theoretically, the proposed algorithm merges pre-calculated detection probability information and prior information into the resampled projection data by applying Bayes’ Theorem. In contrast to conventional projection data handling, RESPACE can make the iterative reconstruction isolated from any detection model or PSF modelling, ensuring the closed structure of normal non-PSF iterative algorithms. In this study, we implemented a 2D-PET simulation Monte Carlo framework, which has the same geometrical property as Siemens BrainPET transverse structure. Conventional MLEM, MLEM-PSF (both image space and projection space with shift-invariant kernel) and RESPACE are implemented and investigated. As figures of merit, Bias-Resolution curves demonstrate that RESPACE could achieve similar resolution and even better bias suppression performance as the PSF method with a shift-invariant kernel. Moreover, no significant visual difference is observed between images from PSF and RESPACE reconstruction. These results demonstrate that RESAPCE offers equivalent performance as the shift-invariant PSF method and this approach is an alternative resolution modelling method independent from the iterative reconstruction algorithm.