Abstract
Direct synthetic contrast enhancement may entangle global appearance with lesion-relevant structure, so perceptual similarity alone may not reflect lesion-level reliability. We formulate synthetic breast DCE-MRI as an enhancement-field decomposition problem, separating global enhancement formation from image-derived residual refinement. Our Coarse-to-Structure (C2S) model predicts a half-resolution coarse enhancement field and performs full-resolution structural refinement, while requiring only a pre-contrast slice at inference. In a three-seed matched warm-start study on 203 local cases, C2S achieved more favorable mean values than the matched single-stage model for MSE,tumor-ROI SSIM, FRD, both AUROCs, Dice, and HD95; the single-stage model achieved lower LPIPS. Case-level bootstrap supported the MSE and Dice differences favoring C2S and the LPIPS difference favoring single-stage. These results indicate a metric-dependent fidelity–structure trade-off within the evaluated setting. This distinction matters when synthesized images are used for lesion-focused downstream analysis, suggesting that perceptual similarity should complement, rather than replace, lesion-level structural evaluation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MAMA_SYNTH_011.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=9LzxQoRic6
BibTex
@InProceedings{YaoYuj_TwoStage_MICCAISAT2026,
author = { Yao, Yujie AND Wang, Guotai},
title = { { Two-Stage Enhancement-Field Synthesis of Contrast-Enhanced Breast MRI from Pre-Contrast Slices } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17256},
month = {pending},
page = {pending}
}