Fernandez-steel skew normal mixture model using bayesian approach for MRI-based brain tumor segmentation

Detection of a brain tumor in Magnetic Resonance Imaging (MRI) is always challenging due to the gray level comparison of tumor and nor-mal tissue. Model-based clustering with a Finite Mixture Model (FMM) is widely used to segment the tumor as the Region of Interest (ROI). The Gaussian Mixture Model (GMM) is becoming abandoned because, in reality, the symmetric distribution approach is less able to explain the MRI data pattern. In addition, the use of a symmetric distribution can-not compete for the model parsimonious of an asymmetric distribution to exhibit the long and heavy tail pattern of the data. On this kind of data, more Gaussian mixture components are needed in the GMM. This study, therefore, develops a mixture model with asymmetric dis-tribution, called Fernandez-Steel Skew Normal (FSSN). It is one of the Neo-Normal distributions that can be skewed adaptively but remains stable in its mode of distribution. Bayesian coupled with the Markov chain Monte Carlo (MCMC) approach is employed for estimating FSSN distribution parameters numerically. Silhouette Index (SI) coecient is performed to validate the result of the segmentation. The results indicate that the FSSN mixture model (FSSN-MM) has a better performance at representing the data pattern of a brain tumor MRI. This is indicated by the higher SI coecient of the FSSN-MM than GMM. In addition, the FSSN-MM is more parsimonious, since it has the smallest number of clusters. Moreover, FSSN-MM is able to detect the brain tumor more precisely than the original GMM approach.