Optimized sampling patterns for practical compressed MRI

The performance of compressed sensing (CS) algorithms is dependent on the sparsity level of the underlying signal, the type of sampling pattern used and the reconstruction method applied. The higher the incoherence of the sampling pattern used for under-sampling, less aliasing will be noticeable in the aliased signal space, resulting in better CS reconstruction. In this work, based on point spread function (PSF) properties, we compare random, Poisson disc and constrained random sampling patterns and show their usefulness in practical compressed sensing applied to dynamic cardiac magnetic resonance imaging (MRI).