Leveraging unsupervised image registration for discovery of landmark shape descriptor

•Proposes an end-to-end model to obtain shape descriptor from images, alleviating pre and post-processing needs.•Self-supervised neural architecture model that leverages image registration to facilitate landmark discovery.•Frameworks provides model variants to encode additional shape information,regularization and heuristic for removal of redundant landmarks.•Discovered landmarks are usable on 2D and 3D images for different downstream tasks ranging from clustering to severity quantification.