Real-time cardiac MRI using low-rank and sparsity penalties

We introduce a novel algorithm to reconstruct real-time cardiac MRI data from undersampled radial acquisitions. We exploit the fact that the spatio-temporal data can be represented as the linear combination of a few temporal basis functions. The current approaches that capitalize this property estimate the basis functions from central phase encodes, acquired with a fine temporal sampling rate. In contrast, we estimate the basis functions from the entire under-sampled data. By eliminating the need for training data, the proposed method can achieve potentially high acceleration factors. More importantly, the estimation of the temporal functions from the entire data significantly improves the quality of the basis functions, which in turn improves the quality of the reconstructions. Experiments on numerical phantoms show a significant reduction in artifacts at high acceleration factors, in comparison to current schemes.