1: \begin{abstract}

2: In this paper we present a fast and efficient method for the reconstruction of Magnetic Resonance Images (MRI) from 

3: severely under-sampled data. 

4: From the Compressed Sensing theory we have  mathematically modeled the problem as a constrained minimization problem

5: with a family of  non-convex regularizing objective functions depending on a parameter  and a  least squares data fit constraint. 

6: We  propose a fast and efficient algorithm, named Fast NonConvex Reweighting (FNCR) algorithm,  

7: based on an iterative scheme where the non-convex problem is approximated by its convex 

8: linearization and the penalization parameter is automatically updated. 

9: The convex problem is solved by a Forward-Backward  procedure, where the Backward step is performed 

10: by a Split Bregman strategy. Moreover, we propose a new efficient iterative solver for the arising linear systems. 

11: We prove the convergence of the proposed FNCR method.

12: The results on synthetic phantoms and real images show that the

13: algorithm is  very well performing  and computationally efficient, even when compared to the best performing methods proposed in the literature.

14: \end{abstract}

15: