Software architecture of the MOLAR-HRRT reconstruction engine

The Motion-compensation OSEM List-mode Algorithm for Resolution-recovery/Reconstruction (MOLAR) is a complete system for managing and performing iterative PET reconstructions. Although it was designed for use with the ECAT HRRT, its pluggable component architecture is readily extendable to any PET scanner, list-mode or frame-mode. List-mode data are stored on a distributed set of linked lists of event packets. Each event packet stores all information needed for one coincidence event. The resolution model uses a distance-based approximation and assumes separability of the radial and axial factors. Since the complete computation of the global sensitivity image Q (the denominator outside the sum in the OSEM update equation) requires a back-projection of all 4.5×10⁹ possible LORs in the HRRT at all time increments in the scan frame and is thus intractable, we have devised an approximation in which the back-projection is performed on only a randomized sub-sampling of LOR space. To test the randomized approach we generated 10 realizations of the Q image at various count levels and measured the average coefficient of variation (COV) per slice at each count level. We characterize the dependence of image noise on error in Q. We also provide a result demonstrating that we have achieved image resolution of less than 3 mm, a critical design goal.