Medical Image Registration

Medical image registration finds transformations that bring corresponding anatomy into alignment, enabling comparisons across time, modalities, or patients and supporting tasks such as motion correction and surgical guidance. Methods range from intensity- and feature-based optimization to deep networks, implicit deformation models, pretrained image features, and geometric or biomechanical priors; transformations may be rigid, affine, or deformable and diffeomorphic. Researchers evaluate these methods on settings such as brain MRI, lung CT, cardiac imaging, and intraoperative 2D–3D alignment, using landmarks, anatomical overlap, image similarity, and deformation quality. A central challenge is balancing accurate local matches with physically plausible, reliable deformations that generalize across scanners and modalities, since metrics can disagree and better overlap does not always mean better anatomical correspondence.