Expectation Maximization And Gaussian Model Based Segmentation on Histology Slides

Abstract

The importance of the Expectation Maximization (EM) algorithm isincreasing day by day in order to solve Maximum A Posteriori (MAP) estimation problems and Gaussian Mixture Models (GMMs), which are parametric probability density functions, have become more popular in computerized applications due tothe EM algorithm. This article explains an automatic GMM based image segmentation method for histology cell images. For this purpose, the GMM parameters, which are recomputed iteratively starting with initial values, arecalculated by using the EM algorithm which classifies each pixel into the class withthe largest probability distribution using maximum likelihood. The accuracy of this segmentation algorithm depends on how much close the probabilistic model to the gray level distributions of the input images.