Abstract
Dynamic contrast-enhanced MRI (DCE-MRI) is essential for breast cancer diagnosis but requires gadolinium-based contrast agents that pose safety and accessibility concerns. We present a conditional generative adversarial network for synthesizing post-contrast breast MRI from pre-contrast T1-weighted acquisitions. Our method introduces two key contributions: (1) per-image z-score normalization that eliminates cross-scanner intensity variation, improving perceptual similarity by 25%; and (2) a predicted tumor mask as conditional input that provides the generator with an explicit spatial prior for enhancement localization, improving downstream segmentation Dice by 14%. We additionally isolate the multi-scale enhancement consistency (MSEC) loss inherited from our earlier multi-parametric work and find it redundant once these two components are present. On our internal validation set, the method achieves a tumor-region SSIM of 0.509, a downstream segmentation Dice of 0.624, and a 95th-percentile Hausdorff distance (HD95) of 69.4, demonstrating that tumor-aware conditional synthesis can produce clinically meaningful virtual contrast enhancement.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MAMA_SYNTH_017.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=F8WhTtpcmc
BibTex
@InProceedings{JiaHon_TumorAware_MICCAISAT2026,
author = { Jiang, Hong AND Yang, Zhikai AND Moreno, Rodrigo},
title = { { Tumor-Aware Conditional Pix2PixHD with Per-Image Normalization for Virtual Contrast Enhancement in 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}
}