Deformable Image Registration
Overview
Deformable image registration finds spatial correspondences between images when structures move or change shape, supporting tasks such as longitudinal analysis, image-guided surgery, and radiotherapy dose accumulation. Methods include classical intensity- and feature-based optimization, spline or diffeomorphic deformation models, and learned CNN or transformer systems; newer approaches also use pretrained features, biomechanical priors, and test-time refinement. Researchers assess alignment with measures such as landmark target error and anatomical overlap, while also checking whether deformation fields are smooth, plausible, and useful for downstream tasks. A central challenge is that methods that perform well on one anatomy or modality may not generalize, and better overlap does not always mean reliable correspondences or safe clinical decisions.
Latest papers
77 papers10 papers in the last four weeks, up 43% on the four weeks before. 0.1% of all new papers.