Abstract

Four-dimensional (4D) flow MRI measures three-directional blood velocity over a three-dimensional volume and the cardiac cycle, but its high dimensionality leads to long scan times and requires aggressive k-space undersampling. Reconstruction is particularly challenging because velocity is derived from phase differences between one reference and three flow-encoded acquisitions. As a result, a reconstruction may look accurate in magnitude while still containing phase errors that distort the estimated velocity field. We treat the four encodings as a coupled representation and introduce a physics-unrolled model with explicit multi-coil data consistency. Its phase-coupled state-space (PCSS) regularizer combines a complex-encoding branch with an auxiliary phase-difference branch, while a phase-difference manifold-consistency (PDMC) objective directly supervises velocity magnitude and direction within the vessel. Acquisition prompts condition both branches on acceleration, VENC, and the vessel mask. On CMRx4DFlow 2026 over R∈(10,20,30,40,50), the 10-cascade model outperforms 10-cascade FlowVN and FlowMRI-Net. We further show that increasing the number of cascades to 28 improves the performance of the same architecture in both magnitude and velocity reconstruction.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRxRecon2026_013.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=N5fOYgAWsb

BibTex

@InProceedings{AnvKia_PhaseCoupled_MICCAISAT2026,
        author = { Anvari Hamedani, Kian AND Ebrahiminia, Fatemeh AND Xiao, Yuliang AND Vavasour, Zach AND Chiew, Mark},
        title = { { Phase-Coupled State-Space Reconstruction for Ultra-Accelerated 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}
}


back to top