Lohith Kini1, Larry L. Wald2,3, Elfar Adalsteinsson1,4
1Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA, USA; 2MGH, Harvard Medical School, A. A. Martinos Center for Biomedical Imagin, Charlestown, MA, USA; 3Harvard-MIT Division of Helath Sciences and Technology, MIT, Cambridge, MA, USA; 4Harvard-MIT Division of Health Sciences and Technology, MIT, Cambridge, MA, USA
Advances in graphics card processors have allowed for faster computation time for solving numerical equations that are highly parallelizable. In this study, we present the use of CUDA enabled graphics cards in FDTD simulations for SAR, E1 and B1 computation. The speed benefit is useful if, for e.g., SAR estimation for pTX is necessary for scanning individual subjects. A fast FDTD computation would also significantly speed up iterative optimizations of a coil design over a geometric parameter space. We show that steady state solutions are achieved quickly and that the running time is at least an order of magnitude greater than regular CPU computation.