Despite the facts that automated image analysis has been and continues to be a very active research area and this PI and others have demonstrated that incorporation of knowledge into the process can significantly improve its robustness and accuracy, there is only very limited knowledge applicable to the general problem of designing knowledge-based image analysis systems. This research begins with the hypotheses that (1) A general image analysis system can be developed for families of similar image analysis applications; (2) Knowledge can be automatically or semi-automatically derived from application-specific training sets of manually analyzed example images; (3) General knowledge-based image analysis systems trained on such training sets can offer performance comparable to single-purpose knowledge-based systems specifically designed for the particular task. Since the general problem is quite broad, the specific aims are to (1) Develop a unifying approach to example-based acquisition and representation of knowledge about image edges derived from example images in a variety of border detection applications; (2) Develop a general approach to automated border detection learning utilizing knowledge acquired from examples in this fashion; (3) Assess the applicability of the developed approaches by testing their robustness and accuracy in large data sets of complex images from actual applications. This research, which will be carried out in the context of medical image analysis, such as angiographic images, intravascular and intracardiac ultrasound images, and multidimensional high resolution CT images, will open new routes to automated and practically applicable machine learning strategies in image analysis and image understanding applications.
