Abstract
Deep learning-based reconstructions have shown promising results to accelerate 4D flow magnetic resonance imaging (MRI). However, current state-of-the-art reconstruction methods offer limited usability and interpretability (both critical for clinical adoption) because network training must be performed independently per acceleration factor (R) and data distribution, with an unknown relationship (and possibly missed mutual benefits) between the networks. In this work, we propose FlowMRI-Net++, improving upon state-of-the-art by 1) Combining a modern network architecture that exploits the multidimensionality of 4D flow MRI with supervised losses, 2) Conditioning the network on R and sampling locations of the acquired k-space data, and 3) Zero-shot adaptation (ZSA) via self-supervised test-time optimization. A subset of the CMRx4DFlow2026 dataset was used, comprising 138 aortic 4D flow MRI scans, which were retrospectively downsampled to generate paired data. We demonstrate that FlowMRI-Net++ quantitatively outperforms state-of-the-art and offers a unified solution for reconstruction of R=20-40× without a loss in velocity and magnitude-based performance metrics compared to dedicated networks, additionally offering improved interpretability in terms of the impact of R on data consistency. Finally, we demonstrate that ZSA can quantitatively improve performance and generalizability to out-of-distribution training data (e.g. +0.025±0.004 in structural similarity index measure [SSIM] for R=30, N=6) at the cost of some computational overhead.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRxRecon2026_003.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=UQYFtJod7L
BibTex
@InProceedings{JacLuu_FlowMRINet_MICCAISAT2026,
author = { Jacobs, Luuk AND Kozerke, Sebastian},
title = { { FlowMRI-Net++: A Unified Sampling-Aware 4D Flow MRI Reconstruction Network with Zero-Shot Adaptation } },
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}
}