A Computationally Efficient MASTeR-based Compressed Sensing Reconstruction for Dynamic MRI

Abstract

State-of-the-art compressed sensing based algorithms recover sparse signals from under sampled incoherent measurements by exploiting their spatial as well as temporal structures. A compressed sensing based dynamic MRI reconstruction algorithm called MASTeR (Motion-Adaptive Spatio-Temporal Regularization) has shown great improvement in spatio-temporal resolution. MASTeR uses motionadaptive linear transformations between neighboring images to model temporal sparsity. In this paper, a computationally efficient MASTeR-based scheme is presented that achieves the same image quality but in less time. The proposed algorithm minimizes a linear combination of three terms (ℓ1-norm, total-variation andleast-square) for initial image reconstruction. Subsequently, least-square and ℓ1-norm with ME/MC i.e., motion estimation and compensation are used to reduce the motion artifacts. The proposed scheme is analyzed for breath-held, steady-state-free-precession MRI scans with prospective cardiac gating