Abstract
Cross-modal MRI-histology registration bridges cellular resolution tissue and in-vivo imaging but remains difficult. The modalities share anatomy yet differ in shape, contrast, and resolution; histology is cut into 20-100 μm sections with shrinkage, cutting artifacts, and staining variability, while MRI is volumetric at 0.5-1 mm. Most learning-based pipelines predict the entire warp, often from synthetic deformations; on cross-modal, cross-cohort tasks this amortization encodes a strong deformation prior that transfers poorly. We argue for a more conservative design: learn only the similarity metric, then solve the deformation per pair. A 76k-parameter PatchNCE encoder learns a dense cross-modal cost surface, a lightweight velocity decoder provides an initial deformation condition for a per-pair diffeomorphic optimizer, and scaling-and-squaring integration yields warps that are fold-free in every experiment reported here. On the published BigBrain-MNI AFID protocol our pipeline reaches 1.70 mm mean Target Registration Error (TRE) on 30 fiducials (on par with the 1.77±1.25 mm literature result) and 1.47 mm on 28 fiducials, with no folded voxels. The same encoder transfers zero-shot to two AHEAD ex-vivo brains (MIND distance ≈ 0.07) without tuning. By contrast, the foundation model multiGradICON reaches Dice 0.972 but at 5.47 mm TRE with 3.15% folding; ours is the only deformable method that simultaneously achieves competitive Dice (0.91), sub-affine TRE, and no folding.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_039.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MLMI_039_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=6WRpwYEudI
BibTex
@InProceedings{MisJig_Learn_MICCAISAT2026,
author = { Mistry, Jigneshkumar AND Ram, Keerthi AND Mulay, Supriti AND Neelakantan, Sivathanu AND Sivaprakasam, Mohanasankar},
title = { { Learn the Metric, Not the Deformation: A Small Contrastive Encoder for Cross-Modal Histology-MRI Registration } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}