Lesion Preserving Image Registration with Applications to Human Brains

The goal of image registration is to find a transformation that aligns one image to another. In this paper we present a novel automatically image registration approach for images with structural distortions (e.g. a lesion within a human brain). The main idea is to define a suitable matching energy, which effectively measures the similarity between the images. The minimization of the matching energy is an ill-posed problem. Hence, we add a regularity energy borrowed from linear elasticity theory, which incorporates smoothness constraints into the displacement. The resulting energy functional is minimized by a Levenberg-Marquardt iteration-scheme. Finally, we give a two-dimensional example of these applications.