First order statistics-based features selection for clustering using Gaussian mixture model

Abstract

In this study, a total of 60 brain DW-MRI images belonging to three patients showing health, multiple hemorrhagic areas in the left temporal lobe and ischemic stroke pathologies were analyzed with a Gaussian Mixture Model (GMM) for classification-based clustering. To optimize the clustering analysis and to investigate the performance of the classification, various first order statistical based features such as entropy, energy, kurtosis and skewness were used as distinguishable features. Also, the mixing proportion of the chosen components of GMM were investigated. Experiments are performed on DW-MRI images acquired with two magnetic field gradient values or b-value (b = 500 s/mm² and b = 1000 s/mm²) and for non diffusion-weighted images (b = 0 s/mm²). The experimental results show that GMM classifier together with kurtosis and skewness first order statistics-based features better discriminate between studied classes and can be applied for components identification successfully.