Improving synthesis and analysis prior blind compressed sensing with low-rank constraints for dynamic MRI reconstruction

In blind compressed sensing (BCS), both the sparsifying dictionary and the sparse coefficients are estimated simultaneously during signal recovery. A recent study adopted the BCS framework for recovering dynamic MRI sequences from under-sampled K-space measurements; the results were promising. Previous works in dynamic MRI reconstruction showed that, recovery accuracy can be improved by incorporating low-rank penalties into the standard compressed sensing (CS) optimization framework. Our work is motivated by these studies, and we improve upon the basic BCS framework by incorporating low-rank penalties into the optimization problem. The resulting optimization problem has not been solved before; hence we derive a Split Bregman type technique to solve the same. Experiments were carried out on real dynamic contrast enhanced MRI sequences. Results show that, with our proposed improvement, the reconstruction accuracy is better than BCS and other state-of-the-art dynamic MRI recovery algorithms.