Correcting geometric distortions of Echo Planar Imaging using demons and reversed phase encoding

Introduction Echo Planar Imaging (EPI) sequences are subject to imaging artifacts, caused by subject motion, eddy currents effects and field inhomogeneity distortions (susceptibility) causing a geometric displacement of voxel intensities along the phase encode direction. Inhomogeneity correction is important to obtain an anatomical correct image which can be aligned with structural MR images. To estimate the displacement field that allows correction, we extend an existing technique, based on acquiring a full EPI sequence and one additional EPI image acquired with reversed gradient polarity along the phase encoding direction [2, 6]. The EPI with reversed gradient polarity contains the same intensity information as a corresponding image of the EPI sequence but with distortions causing voxel shifts in the reverse directions. To find the displacement field between two reversely distorted EPI’s, an image registration problem is solved as in [2]. We propose to use the simpler, more efficient Thirion's demons algorithm [3, 4] and suggest a different registration pipeline for obtaining the displacement fields and name it phase reversed demons (PRD). The PRD is compared to two other correction methods that require additional MRI sequences, leading to increased scan time. These are the gradient field map (FM) [1] and the point spread function (PSF) [5]. We compare the three methods applied to five subjects. The results are compared visually and quantitatively by estimating the statistical dependence with a structural T1-weighted image. The quantitative results indicate that the (PRD) approach is competitive by being more similar with the structural image but inspection of regions in subjects also demonstrates individual cases where the other methods are favorable.