Abstract
Gadolinium-based contrast agents (GBCAs) are essential for dynamic contrast-enhanced (DCE) breast MRI but carry risks including nephrogenic systemic fibrosis, gadolinium deposition, and high cost. Virtual contrast enhancement—synthesizing post-contrast images from pre- contrast acquisitions—could reduce reliance on contrast-enhanced acquisitions in selected clinical settings, but many existing methods treat the task as direct image-to-image translation without explicitly modeling the spatially heterogeneous nature of contrast-related change. We propose DecompCE, which decomposes the predicted change into a soft spatial modulation map P and a signed magnitude map M, formulated within a conditional flow matching framework with an endpoint-consistency constraint designed to preserve single-step inference quality. On 1,506 multi-institutional cases, DecompCE maintains global fidelity comparable to TransUNet while significantly improving tumor-region reconstruction and enhancement correlation. Ablations and interventions support the complementary roles of P and M and effective single-step inference.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MAMA_SYNTH_014.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=sZCNjsTBwX
BibTex
@InProceedings{KukHee_Where_MICCAISAT2026,
author = { Kuk, Hee AND Oh, Hyerin AND Jang, Jiwoo AND Abbas, Zeeshan AND Lee, Seung Won},
title = { { Where and How Much: Enhancement-Decomposed Flow Matching for Virtual Contrast-Enhanced Breast MRI } },
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}
}
