Non-Rigid Registration of Medical Images: Theory, Methods and Applications

Medical image registration [1] plays an increasingly important role in many clinical applications including the detection and diagnosis of diseases, the planning of therapy, the guidance of interventions and the follow-up and monitoring of patients. The primary goal of image registration is to find corresponding anatomical or functional locations in two or more images. This has many applications: Registration can be applied to images from the same subject acquired by different imaging modalities (multi-modal image registration) or at different time points (serial image registration). Both cases are examples of intra-subject registration since the images are acquired from the same subject. Another application area for image registration is inter-subject registration where the aim is to align images acquired from different subjects, e.g. to study the anatomical variability within or across populations. While rigid registration has become a widely used tool in clinical practice, non-rigid registration has not yet achieved the same level of clinical acceptance. Much recent progress has been made, however, in developing improved non-rigid registration techniques. In this article we will illustrate some of the advances which have been made over the last decades. We will discuss some of the theoretical aspects of non-rigid registration and describe methods for their implementation. Finally, we will illustrate how common problems in medical imaging, such as motion correction and image segmentation, can be solved using image registration.