Acceleration of the EM Reconstruction Algorithm for PET Images Using Squared Iterative Methods

• Activity Number: 205
• Type: Contributed
• Date/Time: Monday, July 30, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Constantine E. Frangakis*+ and Ravi Varadhan and Christophe Roland
• Companies: Johns Hopkins University and Johns Hopkins University and University of Science and Technology at Lille
• Keywords: EM algorithm ; emission computed tomography ; polynomial extrapolation methods ; convergence acceleration ; ordered-subsets EM ; SQUAREM

Abstract

Positron emission tomography (PET) is a useful medical technology for imaging the activity of an organ. Reconstruction based on statistical models, using maximum likelihood estimation, provides accurate images. However, the time required for reconstruction is prohibitively high for practical applications, due to the slow convergence of EM algorithm used in the reconstruction. Here we demonstrate the effectiveness of a recently developed class of numerical schemes, called the squared iterative methods (SQUAREM) (Varadhan and Roland, 2004; Roland and Varadhan, 2005), in accelerating the convergence of PET image reconstruction using simulations on a phantom with 128 x 128 pixels. We also compare the performance of SQUAREM to that of the ordered subsets EM (OS-EM) method (Hudson and Larkin, 1994), which is a widely-used, fast reconstruction algorithm.