Real-time dynamic MRI using parallel dictionary learning and dynamic total variation

One of the technical challenges in dynamic magnetic resonance imaging (dMRI) is to obtain MR images with high spatiotemporal resolution in a short scan time. Current state-of-the-art recovery algorithms exploit both spatial and temporal sparsity in dMRI to improve the reconstruction quality. In this paper, we proposed a novel algorithm based on 4-frame parallel dictionary learning and dynamic total variation (PDLDTV) for real-time dMRI reconstruction. The dMRI sequence was decomposed into 4-frame subsets and each subset included the first frame (obtained with more k-space sampling for reference to later frames) and any adjacent three frames. A 3D patch-based dictionary learning algorithm and a dynamic total variation algorithm were used to exploit the spatiotemporal sparsity in each subset. High algorithm speed was required for our real-time reconstruction, and a primal-dual algorithm was used to solve the challenging problem. Experiments over two cardiac dMRI sequences indicated that the proposed 4-frame PDLDTV showed better reconstruction performance than the state-of-the-art online and offline methods such as DTV, k-t SLR, and DLTG.