Abstract
Multi-field MRI synthesis must change field-dependent im-
age appearance while preserving subject anatomy. In MRIxFields2026
this is complicated by a paired set of only three travelling volunteers.
We propose PSP-LDM, a conditional 3D latent diffusion model trained
with a paired–synthetic–paired curriculum: paired training provides di-
rect cross-field supervision, label-conditioned source randomization in-
troduces anatomy and acquisition variability from 1,873 retrospective
targets, and a final paired stage recalibrates the model on acquired pairs
using identity regularization and a three-resolution 3D SSIM loss. A sin-
gle model handles all requested field directions across all three modalities,
and six-sample averaging reached validation SSIM of 0.911, 0.894, and
0.910 on Tasks 1–3. Compared with the reported one-sample baselines,
averaging improved SSIM and nRMSE in every task but increased LPIPS,
trading perceptual detail for structural agreement.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MRIxFields2026_015.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MRIxFields2026_015_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=A6evqjqvMg
BibTex
@InProceedings{ParCha_MultiField_MICCAISAT2026,
author = { Park, Changseon AND Kim, Doyeon},
title = { { Multi-Field MRI Synthesis with Paired–Synthetic–Paired Latent Diffusion } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17274},
month = {pending},
page = {pending}
}