Abstract

Ultra-quality four-dimensional magnetic resonance imaging (UQ-4D-MRI) combines high-spatial-resolution 3D MRI with respiratory-resolved 4D MRI, but learning-based reconstruction is constrained by the scarcity and heterogeneity of clinical 4D-MRI data. We propose ZeroShot-4DRecon, a patient-data-free zero-shot framework pretrained using 1,000 XCAT-derived 4D phantom instances and 1,000,000 augmented training samples. The framework separates global affine alignment from local deformable motion estimation. In VPRSNet, multi-scale features from a vertebral segmentation branch guide keypoint extraction toward relatively stable anatomy, after which paired keypoints are used for least-squares affine fitting. Clinical evaluation included 61 patients and 69 modality-specific 3D-4D MRI pairs from two centers. Without clinical fine-tuning, ZeroShot-4DRecon outperformed the evaluated zero-shot and iterative baselines on most reported metrics across the complete external cohorts. Relative to the separately reported internal-validation results of the fully supervised DDEM baseline, performance differences were modest and metric-dependent, with DDEM remaining better in the Center 2 T2-weighted setting. These findings support synthetic-phantom pretraining as a promising strategy for zero-shot abdominal UQ-4D-MRI reconstruction, while broader transferability remains to be established.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMMCA2026_002.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMMCA2026_002_supp.pdf

Link to Open Review

Open Review Page: https://openreview.net/forum?id=y5IUFhR8T0

BibTex

@InProceedings{WanPei_ZeroShot_MICCAISAT2026,
        author = { Wang, Peilin AND Qi, Pengyuan AND Zhang, Fan AND Liu, Chenyang AND Wang, Lu AND Wang, Xiang AND Li, Zhichu AND Ni, Yimin AND Liao, Yenpeng AND Deng, Jie AND Yang, Zhiyong AND Xie, Zhaoheng AND Cai, Jing AND Li, Tian},
        title = { { Zero-Shot Ultra-Quality 4D-MRI Motion Reconstruction Framework Based on Augmented Digital Phantom Pre-Training } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17278},
        month = {pending},
        page = {pending}
}


back to top