Deep Manifold Learning for Dynamic MR Imaging

Recently, low-dimensional manifold regularization has been recognized as a competitive method for accelerated cardiac MRI, due to its ability capture temporal correlations. However, existing methods have not performed with the nonlinear structure of an underlying manifold. In this paper, we propose deep learning in unrolling manner MRI on Specifically, fixed low-rank tensor (Riemannian) is chosen strong correlations dynamic signals; reconstruction problem modeled CS-based optimization Following structure, Riemannian gradient descent (RGD) adopted solve problem. Finally, RGD algorithm unrolled into neural network, called Manifold-Net, avoid long computation time and challenging parameter selection. The experimental results at high accelerations demonstrate that proposed can obtain improved compared three conventional (k-t SLR, SToRM k-t MLSD) state-of-the-art learning-based (DC-CNN, CRNN, SLR-Net). To our knowledge, work represents first study unroll iterative procedure networks manifolds. Moreover, designed Manifold-Net provides new mechanism priors should also prove useful fast other imaging problems.