A parallel computing framework for motion-compensated reconstruction based on the motion point-spread function

INTRODUCTION Generalized reconstruction frameworks allowing correction of nonrigid displacements during MR acquisition (inter-view motion) have been proposed recently [1,2]. Their main drawback is their computational complexity which makes them currently impractical for clinical use. Based on the properties of the point-spread function (PSF) describing the occurrence of motion-artifacts, we propose a method for splitting the reconstruction process into several reconstruction tasks of smaller size. This allows easier and more efficient implementation of these algorithms on a parallel architecture such as a cluster of workstation, both in terms of reconstruction time and memory requirement.