A Parallel Computing Framework for Motion-Compensated Reconstruction Based on the Motion Point-Spread Function.

Freddy Odille¹, Philip G. Batchelor², Claudia Prieto², Tobias Schaeffter², David Atkinson¹

Generalized reconstruction algorithms have been proposed in order to correct for artifacts induced by nonrigid motion. However they are very time-consuming because large scale inverse problems have to be solved. Here we propose a technique for splitting the reconstruction into several smaller problems, based on the properties of the point-spread function associated with motion artifacts, which uses the local nature of artifacts (blurring) in the frequency-encoding direction. The method was implemented on a cluster of workstations, and applied to the correction of real motion-corrupted data. Efficient motion correction was achieved, with reconstruction times reduced by an order of magnitude.