7864 Parallel Reconstruction using Patch based K-space Dictionary Learning

Introduction: Recently, a parallel reconstruction technique SAKE has been developed using Singular Value Decomposition (SVD) to impose low rank property without using auto-calibrating signal (ACS) data, which can further improve the result provided by SPIRiT with sufficient sampled ACS area. Although predetermined SVD factor analysis method has good analytical and numerical properties, it has been demonstrated that a learned dictionary can better adapt to acquired data and improve the reconstruction result. In this study, we propose a new patch-based dictionary learning method to estimate the local signal features in k-space and demonstrate its improved performance in-vivo. Theory: Low rank matrix completion can be solved by singular-value thresholding. Assuming