An Inter-Iteration Parallel Algorithm for PET Image Reconstruction

Abstract

Image reconstruction using positron emission tomography (PET) involves estimating an unknown number of photon pairs emitted from the radiopharmaceuticals within the tissues of the patient's body. The generation of the photons can be described as a Poisson process and the difficulty of image reconstruction involves approximating the parameter of the tissue density distribution function. Using the Maximum Likelihood (ML) formulation, a better estimate can be made for the unknown image information. Expectation Maximization (EM) is an iterative method which adopts the ML approach, and is the most suitable for PET image reconstruction especially when compared to analytic methods like the filtered back-projection method (FBP). Despite the prognostication of its promises of better medical diagnoses, EM reconstruction has not been widely used in clinical environments because of its long image reconstruction time and large memory requirement. A parallelization scheme of the iterative method has been suggested to make the reconstruction feasible in applied settings. This research builds upon experimentation that demonstrated promising results in speeding up the algorithm in and between iterations using distributed-memory machines.