Abstract
In this paper, we propose a reconstruction method that integrates Deep Diffusion Image Prior (DDIP) into a Cold Diffusion-based accelerated MRI reconstruction framework. Accelerated MRI reduces scan time by undersampling k-space measurements, which makes the image reconstruction problem ill-posed. Diffusion models are widely used to constrain this inverse problem through learned image prior distributions. CDiffMR formulates k-space undersampling as the diffusion process and achieves high reconstruction accuracy by directly incorporating k-space completion into the prior representation. However, learned priors can suffer from prior mismatch when the target anatomy differs from the training distribution. We integrate DDIP-based adaptive prior correction into the reverse process of CDiffMR. The proposed framework refines the learned prior using the measured signal while preserving the artifact-aware sampling structure of CDiffMR. Experiments on the fastMRI dataset evaluate a knee-to-brain cross-anatomy setting, where a model trained on knee MRI is applied to brain MRI without retraining. The proposed integration improves reconstruction accuracy under 4x acceleration. We further analyze the quality–efficiency trade-off and component contributions, indicating the importance of data-consistent correction for stable test-time adaptation. Accelerated MRI Cold Diffusion Deep Image Prior
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/RIME_014.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=Kmh00MQNfr
BibTex
@InProceedings{FukShu_Adaptive_MICCAISAT2026,
author = { Fukutomi, Shunsuke AND Yokota, Tatsuya AND Hontani, Hidekata},
title = { { Adaptive Prior-Guided Cold Diffusion for Accelerated MRI Reconstruction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17277},
month = {pending},
page = {pending}
}