Hierarchical adaptive local affine registration for respiratory motion estimation from 3-D MRI

Non-rigid image registration techniques are commonly used to estimate respiratory motion. Due to the computational complexity of freeform techniques based on control points, hierarchical techniques have been proposed which successively sub-divide the non-rigid registration problem into multiple locally rigid or affine components. A potential drawback of these techniques is that the image content is not considered during the subdivision process. In this paper, we propose a novel adaptive subdivision technique that attempts to automatically divide the image into areas of similar motion, resulting in more accurate local registrations. We demonstrate our new technique by using it to estimate thoracic respiratory motion fields from dynamic MRI data and compare our approach with non-adaptive local rigid and local affine approaches.