MRI Reconstruction by Learning the Dictionary of Spatialfrequency-Bands Correlation: A novel algorithm integratable with PI and CS to further push acceleration

arget Audience: MR researchers on reconstruction and clinical scientists for fast imaging Purpose: Parallel Imaging (PI) 2 and Compressed Sensing (CS) enable scan acceleration by exploiting data correlation among channels and data sparsity in transform domains respectively. However, the acceleration capability is limited by the channel-encoding capability, noise increase and detail blurring. Dictionary Learning has been proposed as a sparsifying transform to improve MRI reconstruction. In this work, we proposed a new algorithm using Dictionary Learning to further improve reconstruction of MRI by exploiting the correlation between image details in different spatial-frequency bands. The results demonstrated advantages over existing PI-CS algorithms and the proposed algorithm can be integrated with PI and CS for further acceleration and improved reconstruction.