Abstract
CMRx4DFlow2026 Regular Task 1 evaluates accurate 4D Flow MRI reconstruction at 10×–50× acceleration, where image magnitude must be recovered without distorting the relative phase that encodes velocity. We propose VeloSubspace, a training-free alternating reconstruction with a temporal basis shared across four velocity encodings, magnitude-weighted circular regularization of relative phase, and explicit multi-coil data consistency. All parameters are selected on the fully sampled P005 demo case and then frozen. Evaluation uses the disjoint P006 case at five acceleration factors, with three deterministic masks per rate. The comparison includes zero-filled SENSE, a matched classical Global LR-DC baseline, and the official learned FlowVN reference. VeloSubspace reduces average magnitude nRMSE from 0.293 to 0.193 and velocity angular error from 64.53 to 50.62 degrees relative to zero filling. Against Global LR-DC (0.232 and 59.08 degrees), it improves all four organizer metrics in 15/15 paired masks; the advantage holds at every acceleration and in leave-one-rate-out summaries. With iteration count matched, circular phase regularization lowers velocity relative and angular errors by 0.0476 and 4.58° on average across five fixed masks. The effect is consistent across all three tested R=20 masks. A three-step acquired-sample update, selected on P005 and then frozen, also improves all four FlowVN metrics at every tested rate. All seven second-center/second-vendor deployment runs produce finite outputs, and every final update lowers the acquired-sample residual. VeloSubspace provides a phase-aware, reproducible Task 1 reconstruction without learned weights.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRxRecon2026_020.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMRxRecon2026_020_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=YKrFc9hasR
BibTex
@InProceedings{CheWei_VeloSubspace_MICCAISAT2026,
author = { Cheng, Weihao AND Ye, Huihui},
title = { { VeloSubspace: Phase-Aware Training-Free Reconstruction for Highly 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}
}