Multi-grid 3D-OSEM reconstruction technique for high resolution rotating-head PET scanners

A fast statistical iterative image reconstruction algorithm has been developed for high resolution PET scanners based on rotating plane detectors. The proposed technique consists of a multi-grid version of the 3D ordered subsets expectation-maximization (3D-OSEM) algorithm. The associated system matrix is precalculated for the fine grid resolution with Monte Carlo methods. Coarse grid system matrices and subset subdivision are derived at a post-process step. Only the system matrix elements associated to the fourth part of central transverse planes are stored in sparse mode format, using axial and in-plane transaxial symmetries during the reconstruction. The multi-grid proposed technique has been evaluated on 3D sinograms obtained from GATE simulations, reporting better resolution-noise trade-off than SSRB+2D-OSEM and FORE+2D-OSEM algorithms, specially in the axial direction and far from the center of the FOV. The proposed algorithm shows significantly faster convergence rate than single-grid 3D-OSEM when is applied to images with local smoothness property.