Parallel Computing for Optimal Control RF Pulse Design

In this work the inherent spatial parallelism of the full time dependent Bloch equation for optimal control based RF pulse design is exploited to accelerate the optimization by means of parallel computing. The major bottlenecks in the provided MATLAB framework are implemented utilizing the MATLAB executable (MEX) interface using sequential C/C++ code, OpenMP CPU multi-threading and CUDA GPU-computing. The demonstrated implementations lead to a significant reduction in computing time while maintaining the high flexibility of the MATLAB environment. In particular, the CUDA implementation allows for optimization times in the order of a few seconds making real-time optimization and patient-specific design feasible. An evaluation of the generated RF pulses indicates no significant loss in accuracy with respect to the MATLAB implementation.