The aim of this work is to improve the accuracy, robustness and efficiency of the compressed sensing reconstruction technique in magnetic resonance imaging. We propose a novel variational model that enforces the sparsity of the underlying image in terms of its spatial finite differences and representation with respect to a dictionary. The dictionary is trained using prior information to improve accuracy in reconstruction. In the meantime the proposed model enforces the consistency of the underlying image with acquired data by using the maximum likelihood estimator of the reconstruction error in partial $k$-space to improve the robustness to parameter selection. Moreover, a simple and fast numerical scheme is provided to solve this model. The experimental results on both synthetic and in vivo data indicate the improvement of the proposed model in preservation of fine structures, flexibility of parameter decision, and reduction of computational cost.

This work aims at developing a generalizable Magnetic Resonance Imaging (MRI) reconstruction method in the meta-learning framework. Specifically, we develop a deep reconstruction network induced by a learnable optimization algorithm (LOA) to solve the nonconvex nonsmooth variational model of MRI image reconstruction. In this model, the nonconvex nonsmooth regularization term is parameterized as a structured deep network where the network parameters can be learned from data. We partition these network parameters into two parts: a task-invariant part for the common feature encoder component of the regularization, and a task-specific part to account for the variations in the heterogeneous training and testing data. We train the regularization parameters in a bilevel optimization framework which significantly improves the robustness of the training process and the generalization ability of the network. We conduct a series of numerical experiments using heterogeneous MRI data sets with various undersampling patterns, ratios, and acquisition settings. The experimental results show that our network yields greatly improved reconstruction quality over existing methods and can generalize well to new reconstruction problems whose undersampling patterns/trajectories are not present during training.

In this paper, we propose a unified variational framework for noise removal, which uses a combination of different orders of fractional derivatives in the regularization term of the objective function. The principle of the combination is taking the order two or higher derivatives for smoothing the homogeneous regions, and a fractional order less than or equal to one to smooth the locations near the edges. We also introduce a novel edge detector to better detect edges and textures. A main advantage of this framework is the superiority in dealing with textures and repetitive structures as well as eliminating the staircase effect. To effectively solve the proposed model, we extend the first-order primal dual algorithm to minimize a functional involving fractional-order derivatives. A set of experiments demonstrates that the proposed method is able to avoid the staircase effect and preserve accurately edges and structural details of the image while removing the noise.