Preliminary examination of the use of case specific medical information as "prior" in Bayesian reconstruction

In this paper we attempt to answer one specific question regarding the role of "prior" distributions in Bayesian reconstructions in Emission Tomography. The question is: Can prior information on some areas of an imaging field improve the results of a reconstruction in other areas of the same image? We answer the question for the specific case of a simple image with a large square of activity surrounding a smaller square, with different ratios of activity in the two regions. For the case of ratios of 10:1 we find that feasible MLE reconstructions exhibit some effect in the internal region due to the presence of the external region and that those effects can be reduced by a factor of ∼0.5 by using as prior information the known mean of activity in the large outer region. Two methods of obtaining reconstructions with pixel dependent constraints are developed in order to obtain the above results. The conclusion of the study is, principally, that "priors" should have a local action, i.e., should apply directly to the region of interest if a substantial improvement in reconstruction quality is to be achieved.