Bayesian reconstruction in emission computerized tomography

Publisher Website
Google Scholar
Abstract
The authors have recently proposed a Bayesian image processing (BIP) formalism and derived iterative BIP algorithms using several different classes of a priori source-distribution constraints. This work is extended to include a statistical Bernoulli process of source elements with fixed total number of photon emissions and a maximum-entropy analysis on individual source-element probability distributions. It is also extended to provide for more-general a priori probability distributions of probable source-element strengths in sub-sets. Such a priori source information is formulated in terms of probability density functions of source-element strengths and spatial correlations. The corresponding iterative BIP algorithms are derived by maximizing the Bayesian functions via the expectation-maximization technique and assuming data obey Poisson statistics. Improved results obtained by applying the algorithms to computer-generated data and experimental phantom-imaging data are presented.