Fully probabilistic model for functional analysis of medical image data
V. Smidl · 2004
Physiological factor images and associated time activity curves are inferred in functional medical imaging, using Bayesian techniques. Specifically, the variational Bayes (VB) framework is employed, leading to an iterative scheme of the expectation-maximization (EM) kind. Positivity constraints are imposed to reflect the physiological nature of the problem. Accelerated convergence is achieved by initializing the VB recursions with appropriately rotated principal components. The procedure is used in a real imaging study of the kidneys, isolating the organs effectively.