ABSTRACT
This chapter focuses on two specific application areas that are increasingly important: image analysis and computer assisted diagnosis system design. It shows that the image context can be modeled by a localized mixture model and that the final image segmentation can be achieved by a probabilistic constraint relaxation network. Neural networks, among other approaches, have demonstrated a growing importance in the area and are increasingly used for a variety of biomedical image processing tasks. The chapter provides examples of the application of the framework in analyses of magnetic resonance and mammographic images. Model-based image analysis aims at capturing the intrinsic character of images with few parameters and is also instrumental in helping to understand the nature of the imaging process. Evaluation of different image analysis techniques is a particularly difficult task, and the dependability of evaluations by simple mathematical measures such as squared error performance is questionable.
