VIOLET
Joint image translation and deformable registration for multimodal medical imaging.
Overview
VIOLET—Volumetric Image registration via Optimization and Learning for Efficient image Translation—reformulates inter-modality registration as an intra-modality problem through image synthesis.
The framework alternates between a three-dimensional diffusion-based image-translation model and diffeomorphic registration. This joint optimization encourages geometrically faithful synthesis and improves alignment across images with substantially different contrasts without requiring perfectly paired training data.