Reconstruction of highly under-sampled dynamic MRI using sparse representation of 1D temporal snippets

This paper introduces a new empirical model for dynamic MRI and shows its application to reconstruction of highly under-sampled dynamic MRI. The model proposes that short 1D signals, so-called snippets, along the image´s temporal dimension are sparse under non-linear transformation using a compact dictionary trained on the data itself. We employ this model to the problem of reconstructing dynamic abdominal MRI and validate its efficacy on a dynamic computational phantom and on an in vivo dynamic MRI sequence. We show how the approach extends and outperforms a state-of-the-art reconstruction algorithm.