Statistical analysis of sheep CAT scan images using a Bayesian mixture model

Abstract

CAT scanning techniques are available to provide images that can aid in the assessment of carcass traits in live sheep during the course of animal experiments. In this paper we present a Bayesian formulation of an analysis that allows us to determine the composition of a scan in terms of proportions of the image attributable to fat, muscle (lean tissue), and bone. The technique, known as finite mixture modelling, also provides information about the distributional properties of some of these components, such as fat and bone. In the case of muscle, the analysis estimates several Gaussian distributions that combine to provide an approximation to its likelihood. The model is estimated through the use of the Gibbs sampler, with the distributional properties of carcass components being obtained from the resultant Markov chains.