Parallel Solvers for Finite-Difference Modeling of Large-Scale, High-Resolution Electromagnetic Problems in MRI. (22nd May 2008)

• Record Type: Journal Article
• Authors: Wang, Hua; Trakic, Adnan; Liu, Feng; Li, Bing Keong; Weber, Ewald; Crozier, Stuart
• Other Names: Ibrahim Tamer Academic Editor.
• Is Part Of: International journal of antennas and propagation. Volume 2008(2008)
• Publication Date: 2008-05-22
• Subjects: Electronic apparatus and appliances -- Periodicals; Antennas (Electronics) -- Periodicals; Radio wave propagation -- Periodicals; Antennes (Électronique); Ondes radioélectriques -- Propagation; Antennas (Electronics); Electronic apparatus and appliances; Radio wave propagation; Electronic journals; Periodicals; 621.382405
• DOI: 10.1155/2008/259703
• Languages: English
• ISSNs: 1687-5869

Abstract

Abstract : With the movement of magnetic resonance imaging (MRI) technology towards higher field (and therefore frequency) systems, the interaction of the fields generated by the system with patients, healthcare workers, and internally within the system is attracting more attention. Due to the complexity of the interactions, computational modeling plays an essential role in the analysis, design, and development of modern MRI systems. As a result of the large computational scale associated with most of the MRI models, numerical schemes that rely on a single computer processing unit often require a significant amount of memory and long computational times, which makes modeling of these problems quite inefficient. This paper presents dedicated message passing interface (MPI), OPENMP parallel computing solvers for finite-difference time-domain (FDTD), and quasistatic finite-difference (QSFD) schemes. The FDTD and QSFD methods have been widely used to model/analyze the induction of electric fields/currents in voxel phantoms and MRI system components at high and low frequencies, respectively. The power of the optimized parallel computing architectures is illustrated by distinct, large-scale field calculation problems and shows significant computational advantages over conventional single processing platforms.