Efficient implementation of the EM algorithm for mammographic image texture analysis with multivariate Gaussian mixtures

In this paper we present an efficient implementation of the EM algorithm for estimating multivariate gaussian mixture model parameters in the context of local-neighborhood image texture analysis. We illustrate its application in a study case of mass detection in mammography, providing a detailed description of a feasible and efficient implementation. Our proposed method overcomes numerical variable underflow problems by means of logarithmic and exponential manipulations and saves computational time using a look up table approach. We reduced computation time to 57.14% with respect to direct computation, achieving numerical conditions for convergence.