Accelerating dynamic cardiac imaging based on a dual-dictionary learning algorithm

Abstract

Traditional CS with dictionary learning (DL) algorithm can be applied in reconstruction for dynamic cardiac imaging (DCI), which is realized by multi-slice two-dimensional format (2D-DLDCI) or directly three-dimensional format (3D-DLDCI). It was reported that dual-dictionary learning algorithm can improve the reconstruction quality for the 3D magnetic resonance imaging (MRI) by introducing prior information and inter-frame correlation. In this study, dual-dictionary learning algorithm was applied in dynamic cardiac imaging (Dual-DLDCI) by exploring the symmetry of the cardiac cycle. High resolution dictionary was trained from the fully acquired previous frames within a period of relaxation, and low resolution dictionary was trained from the under-sampled frames. The patches for traditional 2D dictionary were replaced by the blocks to utilize the spatial correlation among frames. The high resolution dictionary instead of low resolution dictionary was used in the iterative reconstruction to provide prior information. The simulation and experiment results showed that, the Dual-DLDCI algorithm achieves much better reconstruction quality than the other two algorithms.