A Combination of Nonconvex Compressed Sensing and GRAPPA (CS-GRAPPA)

Andre Fischer^1,2, Nicole Seiberlich³, Martin Blaimer¹, Peter Jakob^1,2, Felix Breuer¹, Mark Griswold³

An extension of the nonconvex Compressed Sensing (CS) algorithm is presented. We propose to add a GRAPPA reconstruction step to take advantage of the inherent coil sensitivity profiles of multi-coil datasets. This parallel imaging step introduces a new constraint to the CS algorithm which ensures that the reconstructed data are in accordance with the inherent coil sensitivity profiles. By adding the GRAPPA step, one can reach higher acceleration factors with the CS-GRAPPA algorithm than with each individual technique due to synergy effects of both methods. We were able to recover high quality reconstructions of the sparse dynamic differences between a temporal average and a single timeframe from only 16 interleaved radial projections out of 224 of a dynamic cardiac radial dataset. The results show a significantly improved reconstruction quality compared to nonconvex CS reconstructions without the GRAPPA step. Thus, the proposed method shows potential for all applications where high spatial and high temporal resolution is required.