Calibrationless parallel imaging reconstruction based on structured low-rank matrix completion

• DOI: 10.1002/mrm.24997
• Publication Date: 2013-11-19T02:08:55Z
• Authors: Peter J. Shin; Peder E. Z. Larson; Michael A. Ohliger; Michael Elad; John M. Pauly; Daniel B. Vigneron; Michael Lustig
• Keywords: Matrix Completion; Rank (graph theory); Matrix (chemical analysis); Singular value; Synthetic data

Purpose A calibrationless parallel imaging reconstruction method, termed simultaneous autocalibrating and k-space estimation (SAKE), is presented. It a data-driven, coil-by-coil method that does not require separate calibration step for estimating coil sensitivity information. Methods In SAKE, an undersampled, multichannel dataset structured into single data matrix. The then formulated as low-rank matrix completion problem. An iterative solution implements projection-onto-sets algorithm with singular value thresholding described. Results Reconstruction results are demonstrated retrospectively prospectively Cartesian having no signals. Additionally, non-Cartesian Finally, improved image quality by combining SAKE wavelet-based compressed sensing. Conclusion Because of information needed, the proposed could potentially benefit MR applications where acquiring accurate limiting or possible at all. Magn Reson Med 72:959–970, 2014. © 2013 Wiley Periodicals, Inc.