Abstract
Variabilityinmagneticresonance(MR)imagecontrastacross scanners leads to inconsistent quantitative measurements which prevents reliable biomarker development for brain disorders. Existing harmonization methods often rely on traveling subjects, paired multicontrast images, or synthetic augmentations; thereby restricting their applicability to unseen domains. Such methods rely on inductive biases, such as contrastive disentanglement objectives, yet lack any principled mechanism for incorporating modality-specific prior knowledge for harmonization, leaving the problem significantly underconstrained. We propose a zeroshot MRharmonizationmethodbasedonstructure-constrainedsampling from a phase-preserving diffusion prior. While sampling along the phasepreserving manifold of the source image, our method refines its latent representation through controlled low-frequency injection of the target contrast. Instead of standard Gaussian noise, we train our diffusion prior using structured noise obtained from T -w MR images. For training 1 our prior, we pooled T -w images from three different studies being 1 independently conducted at 47 clinical sites throughout the world. We evaluatedourmethodontwotraveling-subjectsdatasets:anindependent, non-overlapping substudy cohort and the fully out-of-distribution (OOD) FTHP dataset. Our method performed significantly better than recent state-of-the-artanatomy-contrastdisentanglementapproacheswitha10% increase (p < 0.001) in PSNR and a 2% increase (p < 0.001) in SSIM. The source code and pretrained model weights are publicly available at: https://github.com/MFaizyabAli/zshot-phase-harm.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SASHIMI_039.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=ASk7RHULmp
BibTex
@InProceedings{ChaMuh_Sample_MICCAISAT2026,
author = { Chaudhary, Muhammad F. A. AND Remedios, Samuel W. AND Bartz, Kathleen M. AND Hays, Savannah P. AND Zuo, Lianrui AND Koch, Carolyn AND Cassard, Sandra D. AND Wei, Shuwen AND Newsome, Scott D. AND Mowry, Ellen M. AND Dewey, Blake E. AND Prince, Jerry L. AND Carass, Aaron},
title = { { Sample Structure, Refine Contrast: Zero-Shot MR Image Harmonization with a Phase-Preserving Diffusion Prior } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}