Jose Caballero1,
  Anthony Price2, Daniel Rueckert1, Joseph V. Hajnal2
 1Department
  of Computing, Imperial College London, London, United Kingdom; 2Division
  of Imaging Sciences and Biomedical Engineering Department, King's College
  London, London, United Kingdom
The reconstruction of MR data from undersampled observations has been studied as a solution to the acceleration of MR acquisition and shown to have great potential. Nonetheless, exploration of sparsity models has been somewhat limited, particularly in the case of dynamic MRI. Here we propose a combination of dictionary learning techniques and temporal gradient sparsity for the reconstruction of cardiac cine sequences. A comparison with an established method enforcing x-f support sparsity shows the benefits of carefully choosing a model. The technique presented is able to reconstruct the full complex data with an independent treatment of real and imaginary components.