Correction of Dynamic Intensity Inhomogeneity in MR Images

In magnetic resonance imaging (MRI), image intensity inhomogeneity hampers the quantitative analysis, e.g. segmentation or registration. Intensity inhomogeneity is an effect perceived as a smooth variation in intensities across the image. This study provides an efficient, fully automated intensity inhomogeneity correction method that makes no a priori assumptions on the image intensity distribution and is able to correct intensity inhomogeneity with high dynamics. The proposed iterative framework enables incorporation of intensity as well as spatial image features, which are used to compute the force in each image point so that probability distribution of features will condense along the intensity feature. Extensive regularization of forces produces a smooth inhomogeneity correction estimate, which is gradually improved in each iteration. The method was tested on simulated and real MR images for which gold standard segmentations were available. The results showed that the method was successful on all images, having either none, lower or even highly dynamic intensity inhomogeneity.