FAST AND ROBUST GAUSSIAN MIXTURE MODEL FOR MRI BRAIN IMAGE SEGMENTATION

Image segmentation is crucial and preliminary stage of almost all medical imaging diagnosis tools. Gaussian Mixture Model (GMM) is one of common methods for image segmentation and usually, Expectation Maximizing (EM) is used to estimate the parameters of this model. In order to improve EM performance in presence of noise, an extension for EM is proposed which incorporates mean-filtered image as neighborhood information in clustering. In addition, the histogram of image is used as input for clustering to speed up the process. Proposed algorithm quantitatively evaluated in compare to current extensions for EM.