PEM-PET Image Reconstruction in a Clinically-Relevant Time Frame

A positron emission mammography-positron emission tomography (PEM-PET) system is being constructed for PET guidance of breast core biopsy. The system will have two orthogonal pairs of rotating planar detectors operating in coincidence mode, for which performing fully 3D image reconstruction in a clinically-relevant time frame is a challenge. This contribution describes a framework for fast list-mode ordered subsets expectation maximization (OSEM) iterative image reconstruction and two additional techniques to speed image reconstruction. First, a novel analytic approach to sensitivity normalization matrix calculation was developed. Second, parallelization was accomplished for sensitivity normalization matrix calculation and the iteration step of OSEM. The algorithm was tested using a Monte Carlo simulation of a 3 minute clinical PEM-PET acquisition for an ellipsoidal breast phantom with spherical tumors. Timing tests were made on two workstations, one with 2 CPUs and one with 4 dual core CPUs (8 effective processors). On the 8 processor system the sensitivity normalization matrix was computed in 21 sec and fully 3D image reconstruction with this precomputed sensitivity matrix was achieved in 4.6 min. Thus fully 3D iterative image reconstruction for PEM-PET in a clinical time frame is achievable.