Accelerated Perfusion MRI

Union of subspaces and the self expressiveness prior

We assume that the temporal signal at every pixel lies in a union of low dimensional subspaces. We exploit the redundancies present in the data by expressing the temporal signal at each pixel as a sparse weighted linear combination of other temporal signals. This results in a prior that enforces a data point in a subspace to be represented only by points from its own subspace.

We propose an iterative algorithm that alternates between the update of a weight matrix and the dynamic MR images. To further reduce the memory demand and computational complexity, we enforce additional constraints that selects only few points from a small neighborhood in the sparse representation of a data point. We validate the algorithm on a myocardial perfusion MR data and show improvements over state-of-the-art methods such as Low rank and Blind Compressed Sensing (BCS) methods.

References:

Accelerated dynamic MRI using self expressiveness prior.

A. Balachandrasekaran and M. Jacob. ISBI 2015.

Google Sites

Report abuse