Gaussian surface ensemble-based intensity inhomogeneity correction in MR images

Abstract

Intensity inhomogeneity or intensity non-uniformity (INU) is an undesired phenomenon that affects performance of methods both for MR image segmentation and registration. Various techniques have been proposed to eliminate or compensate the INU, most of which are surface fitting algorithms over the entire region. This paper proposes an ensemble of multiple Gaussian surfaces for estimation of INU and subsequently correction in MR images. The individual Gaussian surface is estimated independently over the different homogeneous regions by considering its centre as the centre of mass of the respective homogeneous region. The INU is modeled as a slowly varying multiplicative noise along with the actual tissue signals. Image histogram is considered to extract potential homogeneous regions and then for each of these regions a Gaussian surface is fitted based on the pixel gradients to estimate its INU field or bias field. These bias fields are then ensembled to obtain entire INU field within the MR image. The INU field or bias field is then iteratively removed to obtain the INU-corrected image. The experiments using 2-D synthetic phantoms and real MR images show, that the proposed method performs quite satisfactorily.