The use of mutual information and joint entropy for anatomical priors in emission tomography

This paper studies the use of mutual information and joint entropy to define anatomical priors for maximum-a- posteriori (MAP) reconstruction in emission tomography. Other groups have used mutual information for this purpose, and reported promising results. Simple simulation studies with the "isolated" prior distribution reveal that mutual information may introduce bias, because of a repelling effect between intensity clusters in the marginal histogram. Deleting the terms involving the marginal histograms leads to the joint entropy prior. A gradient ascent MAP-reconstruction algorithm with this prior is described. Its performance is studied with simulation experiments and illustrated on a two sets of patient data: a whole body PET/CT scan and a PET brain scan, combined with the corresponding MRI-scan using off-line registration. The joint entropy seems to be a useful function for defining anatomical priors that do not require explicit segmentation of the anatomical image.