EPDIFF-JF-NET: Adjoint Jacobi Fields for Diffeomorphic Registration Networks

Abstract

This paper presents a deep learning unsupervised
approach for diffeomorphic image registration
called EPDiff-JF-Net. We propose a novel parallel
transport layer to compute the gradients necessary
for training with adjoint Jacobi fields. We test our
method on two independent brain MRI datasets and
obtain state-of-the-art results.