Accelerating Image Reconstruction in Dual-Head PET System by GPU and Symmetry Properties

Cheng-Ying Chou,Weichung Wang,Chin-Tu Chen,Yukai Hung,Yun Dong,Yu-Jiun Kao,Chien-Min Kao,Kewei Chen

doi:10.1371/journal.pone.0050540

Abstract

Positron emission tomography (PET) is an important imaging modality in both clinical usage and research studies. We have developed a compact high-sensitivity PET system that consisted of two large-area panel PET detector heads, which produce more than 224 million lines of response and thus request dramatic computational demands. In this work, we employed a state-of-the-art graphics processing unit (GPU), NVIDIA Tesla C2070, to yield an efficient reconstruction process. Our approaches ingeniously integrate the distinguished features of the symmetry properties of the imaging system and GPU architectures, including block/warp/thread assignments and effective memory usage, to accelerate the computations for ordered subset expectation maximization (OSEM) image reconstruction. The OSEM reconstruction algorithms were implemented employing both CPU-based and GPU-based codes, and their computational performance was quantitatively analyzed and compared. The results showed that the GPU-accelerated scheme can drastically reduce the reconstruction time and thus can largely expand the applicability of the dual-head PET system.

Highlights

Positron emission tomography (PET) is a proven molecularimaging technology for a wide range of biomedical researches and applications [1,2,3,4,5]
Even with the prestored system response matrix (SRM) and exploiting the scanner symmetries to speed up the matrix operations that represent the forward and backward projections, the resulting ordered subset expectation maximization (OSEM) algorithm still requires more than 20 hours to run one iteration [33]; it is difficult to employ the dual-head small-animal PET (DHAPET) scanner for routine small-animal imaging studies
The CPU computational times was, already shortened because there exist a great deal of LORs in the DHAPET system that are shift-invariant or symmetric in some direction

Summary

Introduction

Positron emission tomography (PET) is a proven molecularimaging technology for a wide range of biomedical researches and applications [1,2,3,4,5]. The growing interest in these extended applications inspired development of PET detectors designed for plants [9,10] This is a challenging goal due to the so-called depth-of-interaction (DOI) blurring that leads to reduced image resolution when thick scintillators or compact scanner geometry are used for increasing sensitivity. To address the issue of depth-of-interaction (DOI) blurring, thick detectors [11,12,13,14,15,16,17,18] and accurate image reconstruction methods based on physical and statistical models [19,20,21,22,23] have been developed Despite these extended applications and improvements, high computation cost presents a significant challenge for such adoptions. Each SP has its own Registers to provide the fastest access to a small amount of data and Instruction Units to increase arithmetic density

Objectives

Methods

Results

Conclusion