Abstract
CMRx4DFlow2026 Regular Task 1 reconstructs 4D Flow MRI from 10×–50× undersampled multi-coil k-space. We present one fixed two-stage model. The first stage produces a complex reconstruction and uses the solver residual to select full-strength or regularized data consistency (DC). The second stage is a ten-step untied magnitude refinement model trained across R10–R50. Its loss follows the official acceleration composition and includes a one-sided R50 preservation guard. A positive real gain combines the refined modulus with the frozen production phase. This keeps all cross-encoding phase differences unchanged. On the 20-patient development panel, the DC gate reproduced full DC on 57 of 60 R30–R50 rows and recovered the three high-residual rows from one patient. All-stratum training exposed a low- versus high-acceleration conflict. The guard raised the worst-view SSIM gain to 2.093×10^-3 and the worst-view nRMSE reduction to 1.376×10^-3. An evaluated phase-locked configuration achieved ValidationSet server scores of RelErr 0.241, AngErr 21.043°, SSIM 0.965, and nRMSE 0.033.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRxRecon2026_006.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=mGCYzCXiLS
BibTex
@InProceedings{XiaJia_ResidualGated_MICCAISAT2026,
author = { Xiao, Jiannan},
title = { { Residual-Gated Data Consistency and Phase-Locked Magnitude Refinement for 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}
}