GLOBAL AND LOCAL PRIORS, AND THE LOCATION OF LESIONS USING GAMMA-CAMERA IMAGERY

After a brief review of the paradigm of bayesian image restoration, we pose the question: If high-level prior information is available and usable, what is lost by modelling at the pixel level instead? Our discussion is based on a real application where this question is relevant: the use of gamma-camera imagery in the location of lesions. Procedures using 'global' prior information in the form of a structural model for the image are compared with those using more conventional 'local' priors, modelling only interactions among neighbouring pixels. We address in detail physical modelling and algorithms, and present results with both real and simulated data.