MR image reconstruction from undersampled measurements using union-of-subspaces

This paper presents a new magnetic resonance (MR) image reconstruction method which focuses on estimating the largest K wavelet coefficients. We model the MR image within union of subspaces framework and propose an algorithm named as Subspace Update Algorithm (SUA) to identify subspace. Then we estimate the value of the largest K coefficients by solving the optimization problem consisting of a data fidelity term and a total variation (TV) regularization term. Experimental results on practical MR brain scans show that our proposed method provides high quality reconstruction results. The approach can be applied in many different imaging scenarios.