DiffuseMorph: Unsupervised Deformable Image Registration Using Diffusion Model

Deformable image registration is one of the fundamental tasks in medical imaging. Classical algorithms usually require a high computational cost for iterative optimizations. Although deep-learning-based methods have been developed fast registration, it still challenging to obtain realistic continuous deformations from moving fixed with less topological folding problem. To address this, here we present novel diffusion-model-based method, called DiffuseMorph. DiffuseMorph not only generates synthetic deformed images through reverse diffusion but also allows by deformation fields. Specifically, fields are generated conditional score function between and images, so that can be performed simply scaling latent feature score. Experimental results on 2D facial 3D demonstrate our method provides flexible topology preservation capability.