Consistent region-growing based Dixon water and fat separation for images with disconnected objects

Introduction: Successful water and fat separation in a Dixon technique [1] requires reliable and consistent correction of phase errors that are caused by the magnetic field inhomogeneity. Among different phase correction algorithms that have been published, region growing does not require a priori assumption on the spatial distribution of the field inhomogeneity and has been successful in many applications [2,3]. In a typical region growing scheme, the phase information is successively propagated from one pixel to its neighboring pixels under a general assumption that the magnetic field inhomogeneity is spatially smooth and does not have large discontinuities between the neighboring pixels. A potential problem with region growing is when objects are separated by signal voids. While using a cluster of pixels (e.g. within a boxcar) rather than a single pixel can help region growing overcome small separation between the objects [3], gaps that are well above the boxcar size will essentially break down the information “bridge” and result in region growing failures. As a consequence, the water and fat separation for the disconnected objects within an image will become inconsistent (see Fig. 1). In this work, we demonstrate that modification of a previously-published region-growing algorithm can be used to achieve consistent water and fat separation even when objects are disconnected with large gaps.