LDDMM Meets GANs: Generative Adversarial Networks for Diffeomorphic Registration

Abstract

In this work, we propose an unsupervised adversarial learning LDDMM method for 3D mono-modal images based on Generative Adversarial Networks. We have successfully implemented two models with stationary and EPDiff constrained non-stationary parameterizations of diffeomorphisms. Our approach has shown a competitive performance with respect to benchmark supervised and model-based methods.