Iterative IIR GRAPPA: A Novel Improved Method for Parallel MRI

INTRODUCTION GRAPPA [1] is one of the most popular image reconstruction methods for Parallel MRI. This method interpolates downsampled k-space data with moving average (MA) kernel estimated using a set of fully acquired Auto-Calibrating Signal (ACS) lines in kspace; consequently, its reconstruction quality depends greatly on the estimation and generality of the kernel. Improving kernel estimation and generality in GRAPPA has therefore attracted much research attention recently. An iterative approach called iGARPPA has been proposed in [2] to refine MA kernel estimation and an IIR kernel approach called IIR GARPPA has been introduced in [3] to enhance kernel generality. As we pointed out in [3], while these novel approaches both reuse interpolated data to improve conventional GRAPPA, the IIR kernel approach only reuses these data in auto-regression and has the potential for further improvement if the IIR kernel is refined with the iterative approach. Based on this observation, we propose an Iterative IIR GRAPPA (IIR iGRAPPA) method that uses the iterative approach to refine IIR kernel estimation and hence achieves an image quality superior to that of GRAPPA, iGRAPPA and IIR GRAPPA.