Abstract
Implicit neural representations (INRs) have recently emerged as a promising alternative to population-based CNNs and classical op-timization-based methods for medical image registration, offering continuous, resolution-agnostic deformation fields. However, their instance-wise optimization can be prone to instability and implausible deformations.
Generalizable INRs address this limitation by conditioning a shared network on subject-specific latent priors, yet their smoothness arises implicitly from the network rather than from an explicit constraint, offering no control over the regularity of the deformation.
Moreover, existing generalizable INR frameworks typically require two independent latent priors to represent forward and backward deformations, although both describe the same underlying anatomical correspondence.
We propose \emph{Mirror and Map} (MnM), a symmetric, generalizable INR framework that encodes both deformation directions through a single latent prior and its negation. MnM combines a Lipschitz regularization scheme adapted to the modulated architecture with cycle-consistency regularization and Muon as an optimizer, jointly promoting smooth, plausible, and approximately diffeomorphic deformations. Compared with generalizable baselines, MnM improves registration accuracy while preserving deformation plausibility. We further show that the learned self-supervised deformation representation achieves classification performance comparable to image-based approaches. Code will be available at \url{https://github.com/bkahrs/MirrorAndMap}.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/Off_Grid_037.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=ENas4m1qsB
BibTex
@InProceedings{KahBen_Mirror_MICCAISAT2026,
author = { Kahrs, Bennet AND Falta, Fenja AND Ehrhardt, Jan AND Handels, Heinz AND Kepp, Timo},
title = { { Mirror and Map: Symmetric Latent Deformation Priors for Deformable Image Registration using Implicit Neural Representations } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17277},
month = {pending},
page = {pending}
}