Fast simulation of E1, B1 and SAR for 7T Magnetic Resonance Imaging with the use of graphical processors

Thesis (M. Eng.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2009.Cataloged from PDF version of thesis.Includes bibliographical references (p. 71-73).Specific Absorption Rate (SAR) is a dominant constraint in high field MR, and has been a topic of much recent interest with developments of parallel transmission systems (pTx). While real-time estimates of local SAR over large volumes as well as SAR-constrained pTx RF design are highly desirable goals, it is both difficult to control and computationally demanding. Steady advances in graphics cards for game developers have enabled dramatic speedups in computationally heavy tasks for computer graphics, and some of this functionality is applicable for faster numerical SAR simulation compared to general CPUs. In this study, we present the use of Compute Unified Device Architecture (CUDA) enabled graphics cards in Finite Difference Time Domain (FDTD) simulations for SAR computation. We show that using this framework can speed up computation by at least an order of magnitude compared to regular CPU computation. This will allow us to estimate SAR, B1, and E1 fields quickly for instances where SAR estimation for parallel transmission imaging of individual subjects is necessary, or for optimizing coil designs based on these estimates. A fast FDTD computation would also significantly speed up iterative optimizations of coil design over a geometric parameter space. A description is provided of how FDTD with Uniaxial Perfect Matching Layer (UPML) boundary conditions was coded on GPUs using the NVIDIA CUDA framework. FDTD equations were CUDA optimized by use of two kernel functions, one for the E field update equations and another for the B field update equations. FDTD simulations were compared to an analytical validation case of a dielectric sphere under a current loop. In addition, a description is provided of how SAR computation was parallelized for the CUDA framework.by Lohith Ganesh Kini.M.Eng