Abstract
We propose CUT-4DFlow, a complex-valued, unrolled transformer for accelerated 4D-flow MRI reconstruction. It extends FlowMRI-Net by replacing its complex-valued sequential convolutional-recurrent layers with complex transformer layers, whose attention is parallelizable. We further augment the FlowMRI-Net loss with a term acting on the velocity-encoding differences in k-space, making it more sensitive to velocity. The network is also set up as a unified pipeline that can be trained in either supervised or self-supervised mode, depending on data availability. Because the objective is defined in k-space rather than in the image domain, it is not tied to a specific anatomy, which can aid generalization. We evaluate CUT-4DFlow on aortic reconstruction and assess its generalization across clinical sites, diseases, and anatomies. The CUT-4DFlow code is available at https://github.com/Abdelrahman912/CUT-4DFlowgithub.com/Abdelrahman912/CUT-4DFlow.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRxRecon2026_004.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=6wrZyBQWmP
BibTex
@InProceedings{FatAbd_CUT4DFlow_MICCAISAT2026,
author = { Fathy, Abdelrahman AND Moulin, Kevin AND Perotti, Luigi E. AND Ogiermann, Dennis},
title = { { CUT-4DFlow for Accelerated 4D-Flow MRI Reconstruction } },
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}
}