Abstract

Magnetic Resonance Imaging (MRI) offers unmatched soft-tissue contrast without ionizing radiation, yet is among the least equitably distributed modalities: long acquisitions limit patient throughput, while modern reconstruction networks can exceed the compute available in low-resource clinics. Accelerated reconstruction can reduce acquisition burden, but many high-performing unrolled and diffusion models are large and trained separately for different acquisition settings. We revisit accelerated MRI reconstruction through the lens of resource efficiency and propose MR-DiffRecon, a unified, single-stage framework that supports highly undersampled acquisitions while reducing the inference footprint. Within each dataset, a single model reconstructs across contrasts, views, and multiple acceleration factors (up to x 10), while the same architectural design is independently validated on a second anatomy using dataset-specific training. Adjacent-slice features are aligned at multiple scales via deformable convolutions and fused with implicitly learned prompts, while a training-time reverse diffusion process improves fidelity without requiring diffusion sampling at inference. MR-DiffRecon achieves the strongest SSIM and PSNR among the evaluated methods on CMRxRecon and FastMRI multi-coil knee, while using 23.49% fewer parameters and 7.7% fewer GFLOPs than PromptMR. Code and models are available at https://github.com/wafaAlghallabi/MR-DiffRecon.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_096.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=uRj2YA0djm

BibTex

@InProceedings{AlWaf_MRDiffRecon_MICCAISAT2026,
        author = { Al Ghallabi, Wafa AND Dudhane, Akshay AND Khan, Salman AND Khan, Fahad Shahbaz},
        title = { { MR-DiffRecon: Efficient All-in-One MRI Reconstruction with Multi-Scale Alignment and Implicit Prompting } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17265},
        month = {pending},
        page = {pending}
}


back to top