Tissue segmentation of MR images using first order polynomial modeling

Many magnetic resonance (MR) brain image segmentation techniques assume that the image is formed by classes of biological tissue having constant intensities. However, the presence of inhomogeneities have proven that they tend not to be so. Earlier techniques that tried to cater to inhomogeneities, specifically the bias field, have shown to require pre-setting of parameters or use prohibitive amount of computational resources. In the present approach, a two-dimensional statistical clustering technique based on Bayesian theory is used to model class intensities. To cater for inhomogeneities, class intensities are modeled as polynomials rather than just constant values. A greedy algorithm based on the Iterative Conditional Modes (ICM) algorithm is used to find an optimal segmentation while the model parameters are estimated. The approach can also be easily extended to three-dimensional information and higher order polynomials. Experiments with phantom and real two-dimensional MR images using first order polynomial showed promising results.