BAYESIAN IMAGE-RECONSTRUCTION IN POSITRON EMISSION TOMOGRAPHY

Recent developments in statistical theory and associated computational techniques have opened new avenues for image reconstruction in positron emission tomography. Algorithms based on maximum likelihood methods have received considerable attention for their potential of providing improved image quality. However, noise properties of those images produced by these methods tend to deteriorate after a certain stage in the iterative process. We have developed a Bayesian method that incorporates a priori information to improve the resulting image quality. This approach utilizes a Gibbs prior to describe correlation of neighboring regions and takes into account the effect of limited spatial resolution. The Gibbs prior includes features depicting the similarity of intensities in neighboring pixels within homogeneous regions and line sites outlined as boundaries between regions. The effect of limited spatial resolution is incorporated into the probability density functions relating image cells to detector bins. Other physical factors, in principle, can be included as well. A highly efficient computational technique, the iterative conditional averages method, was employed for computing the point estimates. Significant improvements in image quality can be anticipated. Copyright © 1990 by The Institute of Electrical and Electronics Engineers, Inc.