| Nonrigid registration of images is an important and often crucial step in many areas of image processing and analysis, yet it is only used in a small percentage of possible applications. Automated registration methods are not sufficiently robust to handle complex deformations and locally deviating intensities. The goal of this project is to develop methodology that learns to cope with such situations, from example registrations defined by experts. Image processing by learning has been successfully applied for image segmentation, but the concept is new to image registration. Besides learning how to deform one image to fit another, the system will also learn to detect misregistered areas in the image. These failed areas will not be actively registered, but instead they will be passively deformed by the surroundings. In order to handle complex deformations in a computationally feasible manner, the deformation grid will be selectively refined. The decision to refine a certain area can also be learnt. The methodology will be evaluated on registration of lung computed tomography (CT) scans. Two separate registration tasks in this application area will be examined: change detection in follow-up scans and registration of inspiration and expiration images. We have a large database as well as expert knowledge from radiologists available. |
