Meeting Banner
Abstract #4228

3D TV-Based Compressed MR Image Reconstruction Using a Primal Dual Algorithm

Abolfazl Mehranian1, Hamidreza Saligheh Rad1, 2, Mohammadreza Ay1, 2, Arman Rahmim3

1Medical Physics and Biomedical Engineering Department, Tehran University of Medical Sciences of Medical Sciences, Tehran,Iran, Tehran, Iran; 2Research Center for Science and Technology in Medcine, Imam Hospital, Tehran, Iran; 3Department of Radiology, School of Medicine, Johns Hopkins University, Baltimore, MD, United States

The compressive sensing (CS) of spirally encoded MR acquisitions makes it possible to significantly reduce the scanning time in 3D MR imaging techniques. In this work, we studied an efficient primal-dual algorithm for 3D total variation (TV) and Huber-based compressed MR image reconstruction. We tailored this algorithm for TV and Huber regularizations in 3D and made use of a stack of variable-density spiral trajectories for 80% k-space undersampling. In a volumetric cardiac dataset, it was demonstrated that the derived algorithm objectively outperforms several state-of-the-art algorithms and thus can have promising clinical implications in fast MR imaging.

Abstract PDF E-Poster Video

Keywords

acquisition additive adopted advanced aliasing anatomical appropriate artifacts attempts backward best better biomedical cardiac compressed consider controls criterion dataset define density derivative derive details developed direct dual efficient enhancement equate equations fact fast features fidelity focus forward frequency gradient gradients hence hereby hospital identity imam improved introduced inversion johns king known lemma make math matrix medical much namely nary noise norm operator original otherwise physics pock primal prior priors problem promising promotes promoting proposed proximal recast reconstructed reconstruction reconstructions recover recovers reduced reduces regularizations salsa samples sampling school science selected selection sensing shrinkage solve space spaces sparsity special spiral split stack system technology term thereby thresholding titles toward trajectories trans trends true unitary variable wise zero