Learning from the expert: Improving boundary definitions in biomedical imagery

Abstract

Defining the boundaries of regions of interest in biomedical imagery has remained a difficult real-world problem in image processing. Experience with fully automated techniques has shown that it is usually quicker to manually delineate a boundary rather than correct the errors of the automation. Semi-automated, user-guided techniques such as Intelligent Scissors and Active Contour Models have proven more promising, since an expert guides the process. This paper will report and compare some recent results of another user-guided system, the Expert's Tracing Assistant, a system which learns a boundary definition from an expert, and then assists in the boundary tracing task. The learned boundary definition better reproduces expert behavior, since it does not rely on the a priori edge-definition assumptions of the other models.