Variational level set approach for automatic correction of multiplicative and additive intensity inhomogeneities in brain MR Images

Retrospective correction of intensity inhomogeneities in magnetic resonance images of the brain is an essential pre-processing step before any sophisticated image analysis task can be performed. A popular choice when defining the degradation model in MR images is to use multiplicative intensity inhomogeneities that slowly varying across the image domain; this approach has been extensively used for bias field estimation. However, such a multiplicative model is often insufficient given that some of the most dominant physical causes of intensity inhomogeneities in MRI (such as nonuniform excitation strength) have a non-linear relationship with the receptor signal intensity. In this study, we consider a linear image degradation model with multiplicative and additive intensity inhomogeneity components. We propose a variational level sets approach that combines estimation of intensity inhomogeneity components during the image segmentation process. The evaluation of proposed approach on real MR image datasets demonstrate accurate estimation of multiplicative and additive intensity inhomogeneities improving brain tissue segmentation.