Fast dynamic MRI via nuclear norm minimization and accelerated proximal gradient

Higher spatial and temporal resolution of dynamic MR imaging can be achieved by sparse and sub-Nyquist sampling of (k-t)-space. However, direct inversion of this inverse problem can result in artefact in reconstructed images. In combination with a golden angle pseudo-radial acquisition, we propose in this paper to use prior information based on the rank to regularize the problem. The iterative scheme to reconstruct the dynamic imaging series is based on an accelerated proximal gradient algorithm designed for large-scale low-rank matrix completion. The method is tested on simulated and clinical datasets and, besides being simple, proves to be fast and efficient for high acceleration factors.