Low rank recovery with manifold smoothness prior: Theory and application to accelerated dynamic MRI

We introduce a regularized optimization algorithm to jointly recover signals that live on a low dimensional smooth manifold. The regularization penalty is the nuclear norm of the gradients of the signals on the manifold. We use this algorithm to reconstruct free breathing dynamic cardiac CINE MRI data. A novel acquisition scheme was used to facilitate the estimation of the manifold structure and recover high quality images. The results show that the method is an efficient alternative to traditional breath-held CINE exams.