A GPU accelerated modeling of bio-effects associated with magnetic resonance imaging

With the recent development of high field MRI scanners, the risk for healthcare staff being exposed to large static magnetic fields (3T to 7T) and rapidly time-varying magnetic field gradients is greatly increased. A better understanding of the interaction mechanisms and the bio-effects associated with MRI environment would allow sensible and workable exposure limits to be set for staff, patients and volunteers. This paper presents a novel approach in modeling hazardous electric field levels induced in a human body under continuous movements within a strong magnetic field environment. The derived algorithm is able to accurately model both translational motion and rotating body movements. Since this algorithm is based on the quasi-static Finite-Difference approximation, the computational space for modeling a human body can then be divided into a large number of cubic cells. Every cell in the model is very suitable for parallelization and hardware acceleration using General Purpose Graphical Processing Units (GPGPU). After adopting several optimization techniques, a speedup of around 40 times is achieved by adopting GPGPU for modeling torso movements around 8 million cells compared with a CPU implementation.