Computer Vision News - December 2022

37 Mohamed Suliman individual subject surface in a way that improves the similarity to the template. The displacements are learned in a discrete manner, taking inspiration from recent deep-discrete registration frameworks, which show capability in learning larger deformations than continuous-based ones . To learn the deformation field on the cortical surface, we convert the registration problem to a multi-label classification problem , where each point in a low-resolution control grid on the cortex deforms to one of fixed, finite number of endpoints on the cortex. This deformation is learned using a spherical geometric deep learning architecture , in an end-to-end unsupervised way, with regularization imposed using a deep Conditional Random Field (CRF) . Our results show that GeoMorph performs competitively in terms of alignment and distortion qualities relative to the most famous classical surface registration algorithms while superbly outperforming them in running time . What is more promising is that GeoMorph is capable of generating smoother deformations than all existing learning-based surface registration methods, even in subjects with atypical cortical morphology, i.e., generating neurobiologically plausible distortions .