Massively parallel CUDA simulations of cardiac and embryonic MRI on a cloud-based cluster

Embryonic and cardiovascular Magnetic Resonance Imaging (MRI) in both clinic and research environments is confronted by challenges like motion, flow and low resolution. MRI simulations can be used as a valuable tool against these challenges but they are characterized by high computational intensity. In this study we implemented cardiac and embryonic MR simulations on a cloud-based cluster. For this purpose, an anatomical model of a human embryo was split over a variable number of computer nodes on the cloud. Simulation of a Gradient-Echo pulse sequence was performed and the partial results were aggregated in a central node to produce the simulated MR image, following the MapReduce paradigm. The measured execution times and speedups demonstrated the benefits of this cloud-based approach.