BAYESIAN ESTIMATION OF SPATIALLY VARIANT FINITE MIXTURE MODEL FOR BRAIN MR IMAGE SEGMENTATION

Abstract

The Finite Mixture Model (FMM) based approaches have been applied in Magnetic Resonance Imaging (MRI) to extract information about human anatomy. The idea is to model feature vector of a tissue using some known distribution (such as Gaussian, known as GMM). The performance of FMM deteriorates with increase in noise within data which may occur due to environment, patient movement, technician expertise level etc. The Spatially Variant Finite Mixture Model (SVFMM) is used as robust alternative to this. The student’s t distribution has been explored, in place of Gaussian distribution, in context of FMM. Within the framework of SVFMM, Student’s t distribution is used in this work to overcome the noise impact present in the data. The parameters of SVFMM are learned by ‘sampling-resampling’ based Bayesian Learning. The novelty of work lies in parameter estimation of mixture model in comparison to Frequency based estimations. This paper compares different mixture models, distribution functions and parameter estimation procedures. The column based sampling is used to train the model. The misclassification rate (MCR) is used as quantitative measure for performance evolution. The results of proposed method seem encouraging.