Abstract
Dynamic contrast-enhanced MRI (DCE-MRI) is central to breast cancer characterization, but the gadolinium-based contrast agents it requires raise safety, environmental, and access concerns. Virtual contrast synthesis aims to recover the diagnostic post-contrast appearance from a pre-contrast acquisition. In practice, generative models tend either to smooth away lesion vasculature or to depend on tumor masks that are not available when the lesion location is unknown. We present DARE, a single-pass residual generator that pairs a Dilated Anatomy-Guidance(DAG) stream with a U-Net whose bottleneck uses shifted-window self-attention. The DAG stream widens the receptive field through cascaded atrous residual blocks while keeping the native resolution, and it feeds anatomical detail into every scale of the backbone. The network predicts the pre-to-post residual and is trained with a region-weighted objective that combines a tumor-aware Charbonnier term, a gradient term, a mask-weighted perceptual term, a multi-scale adversarial and feature-matching term, and an auxiliary segmentation-consistency term. Masks are used only to shape the loss; inference needs no mask. On the multi-center MAMA-SYNTH (MAMA-MIA) cohort, with a source-disjoint validation set used for model selection, DARE reaches a mean SSIM of 0.75, MS-SSIM of 0.83 and PSNR of 25.2 dB on in-domain validation, and holds comparable values on the held-out sources (SSIM 0.72, MS-SSIM 0.83, PSNR 25.5 dB), so the synthesized enhancement transfers to sources excluded from training without a mask at test time.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MAMA_SYNTH_009.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=AlGoXZeNbR
BibTex
@InProceedings{GeoKon_DARE_MICCAISAT2026,
author = { Georgas, Konstantinos AND Vagenas, Theodoros P. AND Matsopoulos, George K.},
title = { { DARE: Dilated Anatomy-Anchored Residual Enhancement for Virtual Contrast Synthesis in Breast DCE-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}
}