The use of constraints to eliminate artifacts in maximum-likelihood image estimation for emission tomography

Abstract

Images produced in emission tomography with unconstrained maximum-likelihood estimation techniques exhibit two artifacts as the likelihood hill is climbed and the images converge toward one with maximum likelihood. The first artifact is a speckled appearance on the image, a noise artifact. The second is an estimation error near edges of the underlying radioactivity concentration. However, if mathematical constraints are used that blur the estimated image and the image being estimated, these artifacts are eliminated. The authors review the nature of these constraints, demonstrate them, and measure the performance of the modified algorithm relative to the classical linear image-estimation algorithm.