Modified iterative model based on data extrapolation method to reduce Gibbs ringing

Abstract

An iterative algorithm that involves image filtering and data replacement (as suggested by Constable and Henkelman) is investigated for reducing the Gibbs artifact in magnetic resonance imaging. The image is processed with an edge-preserving filter to estimate the height and location of a set of model elements (delta functions or boxes) for generating the missing high-frequency information. Filtering was performed in the complex image domain to account for discontinuities in phase as well as magnitude. The process is repeated after each data replacement to handle varying degrees of contrast. The convergence and signal-to-noise characteristics of the algorithm are investigated by means of simulated and clinical examples. The results indicate that the algorithm performs reasonably well in reducing ringing artifacts due to nearby edge contrast seen in most of the homogeneous, isointense regions. Nevertheless, it may generate some spurious thickening of structures that do not match the assumed step-edge models.