Abstract
Four-dimensional flow magnetic resonance imaging (4D flow MRI) enables time-resolved blood flow imaging but requires clinically impractical scan times to resolve multiple cardiac phases and three velocity-encoding directions. Although unrolled network reconstruction has been introduced to accelerate 4D flow MRI, supervised and self-supervised methods exploit limited temporal information and may degrade under acquisition domain shifts. To address these limitations, we propose a subspace-guided upstream-downstream reconstruction framework coupling supervised pretraining with self-supervised, proximally regularized adaptation. A global low-rank spatial subspace is estimated across the entire cardiac cycle and incorporated into each unrolling layer, constraining the 3D CNN regularizer to model the complementary spatiotemporal residual. The upstream stage learns a generalizable prior from fully sampled training data, while the downstream stage adapts the pretrained parameters to each unseen acquisition using only its undersampled measurements. On the CMRx4DFlow2026 aortic dataset (R=10–50), the proposed framework achieved superior magnitude (SSIM, nRMSE) and velocity (RelErr, AngErr) performance compared with CS-LLR, FlowVN, and FlowMRI-Net.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRxRecon2026_012.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMRxRecon2026_012_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=sr9Tc0PpqL
BibTex
@InProceedings{LeeMye_SubspaceGuided_MICCAISAT2026,
author = { Lee, Myeongjin AND Ryu, Hyunseung AND Kim, Hahnsung AND Park, Suhyung},
title = { { Subspace-Guided Upstream-Downstream Deep Learning Reconstruction for Fast 4D Flow MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17267},
month = {pending},
page = {pending}
}