Parallelization of the EM algorithm for 3-D PET image reconstruction

Abstract

The EM algorithm for PET image reconstruction has two major drawbacks that have impeded the routine use of the EM algorithm: the long computation time due to slow convergence and a large memory required for the image, projection, and probability matrix. An attempt is made to solve these two problems by parallelizing the EM algorithm on multiprocessor systems. An efficient data and task partitioning scheme, called partition-by-box, based on the message passing model is proposed. The partition-by-box scheme and its modified version have been implemented on a message passing system, Intel iPSC/2, and a shared memory system, BBN Butterfly GP1000. The implementation results show that, for the partition-by-box scheme, a message passing system of complete binary tree interconnection with fixed connectivity of three at each node can have similar performance to that with the hypercube topology, which has a connectivity of log(2) N for N PEs. It is shown that the EM algorithm can be efficiently parallelized using the (modified) partition-by-box scheme with the message passing model.

Comparison of 3-D maximum a posteriori and filtered backprojection algorithms for high-resolution animal imaging with microPET

“…The physical performance characteristics of microPET have been described in detail elsewhere [1]. Traditionally, 3-D image reconstruction is performed with linear algorithms [2]- [4], but the promise of higher resolution and superior noise performance from algorithms that accurately model the system response and the statistical properties of the data have led to the development of 3-D iterative algorithms [5]- [9]. In this work, we compare results obtained using a 3-D filtered backprojection method (FBP), PROMIS, with those obtained using a fully 3-D maximum a posteriori (MAP) method [9].…”

Image Reconstruction Algorithms in PET

“…Pioneering work proposed the use of a cluster of commodity PCs [97] or dedicated hardware [98]. But as soon as commercial parallel systems became available, dedicated algorithms were developed on high-end computers such as transputers [99,100], hypercubes [101], meshes [102], rings [103], fine-grain message- 88 Positron Emission Tomography passing machines [104], linear arrays of DSPs [105] to cite a few examples. Recent efforts concentrated on using clusters of multi-processor PCs, sometimes called component off the shelf (COS), and combine both shared and distributed memory approaches.…”

SPMD cluster-based parallel 3D OSEM

“…However, as commercial parallel systems became available, attention quickly turned to these architectures. Algorithms were developed and implemented for transputer-based systems [6], [7], hypercubes [8], [9], meshes [10], rings [11], fine-grained SIMD machines [12], and linear [13] and systolic [14] arrays, among others. Somewhat ironically, with the rise of "Beowulf"-style computing [15], recent efforts have been directed toward the development of efficient reconstruction algorithms using networks of commodity PCs or workstations [16], [17], much as suggested in the early proposal [3].…”