Correction of fat-water swaps in dixon MRI

Abstract

The Dixon method is a popular and widely used technique for fat-water separation in magnetic resonance imaging,and today,nearly all scanner manufacturers are offering a Dixon-type pulse sequence that produces scans with four types of images: in-phase,out-of-phase,fatonly,and water-only. A natural ambiguity due to phase wrapping and local minima in the optimization problem cause a frequent artifact of fat-water inversion where fat-and water-only voxel values are swapped. This artifact affects up to 10% of routinely acquired Dixon images,and thus,has severe impact on subsequent analysis. We propose a simple yet very effective method,Dixon-Fix,for correcting fat-water swaps. Our method is based on regressing fat-and water-only images from in-and out-of-phase images by learning the conditional distribution of image appearance. The predicted images define the unary potentials in a globally optimal maximum-a-posteriori estimation of the swap labeling with spatial consistency. We demonstrate the effectiveness of our approach on whole-body MRI with various types of fat-water swaps.