Iterative Estimation of MRI Sensitivity Maps and Image Based on SENSE Reconstruction Method (iSENSE) ANGSHUL MAJUMDAR, RABAB K. WARD Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, BC, Canada V6T1Z4 ABSTRACT: SENSitivity Encoding (SENSE) is a parallel MR image reconstruction technique that yields optimal results when the sensitivity maps are accurately known. Unfortunately, in practical scenarios, obtaining accurate estimates of the sensitivity maps is not possible. In this work, we propose a technique that iteratively reconstructs the image and refines the sensitivity maps (from initial estimates). Our technique is named iSENSE (iterative SENSE). Our proposed technique exploits the sparsity of the MR image in some transform domains or the rank defi- ciency characteristic of the matrix representing the MRI image; the former leads to a com- pressed sensing-based reconstruction method, whereas the latter leads to an image reconstruc- tion method that minimizes the nuclear norm (NN) of the image matrix. The sensitivity maps are assumed to be rank-deficient matrices, and thus the refinement of the sensitivity maps is achieved via the NN minimization. To evaluate the performance of the proposed method, we have carried out the experiments on real and one simulated datasets. We have compared our method with three state-of-the-art image domain methods—SparSENSE (Sparse SENSE), NNSENSE (NN Regularized SENSE), and JSENSE (Joint SENSE reconstruction)—and one widely used frequency domain method—GRAPPA (Generalized Autocalibrating Partially Paral- lel Acquisition). Our method yields the best reconstruction results both in quantitative (normal- ized mean-squared error) and qualitative (visual inspection of reconstructed and difference images) evaluation. Ó 2012 Wiley Periodicals, Inc. Concepts Magn Reson Part A 40A: 269–280, 2012. KEY WORDS: parallel MRI; SENSE; compressed sensing; nuclear norm minimization INTRODUCTION The objective of multicoil parallel MRI is to reduce the data acquisition time. Instead of scanning the whole k-space by a single receiver coil, parallel MRI uses multiple coils; each of the receiver coils par- tially samples the k-space. The sampling mask used to undersample the k-space is the same for all the Received 2 July 2012; revised 22 August 2012; accepted 17 September 2012 Correspondence to: Angshul Majumdar. E-mail: angshulm@ ece.ubc.ca Concepts in Magnetic Resonance Part A, Vol. 40A(6) 269–280 (2012) Published online in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/cmr.a.21244 Ó 2012 Wiley Periodicals, Inc. 269