List of Papers Browse by Subject Areas Author List
Abstract
We introduce MRF-DiPh, a novel physics informed denoising diffusion approach for multiparametric tissue mapping from highly accelerated, transient-state quantitative MRI acquisitions like Magnetic Resonance Fingerprinting (MRF). Our method is derived from a proximal splitting formulation, incorporating a pretrained denoising diffusion model as an effective image prior to regularize the MRF inverse problem. Further, during reconstruction it simultaneously enforces two key physical constraints: (1) k-space measurement consistency and (2) adherence to the Bloch response model. Numerical experiments on in-vivo brain scans data show that MRF-DiPh outperforms deep learning and compressed sensing MRF baselines, providing more accurate parameter maps while better preserving measurement fidelity and physical model consistency—critical for solving reliably inverse problems in medical imaging.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/0264_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/p-mayo/mrf-diph
Link to the Dataset(s)
N/A
BibTex
@InProceedings{MayPer_Physics_MICCAI2025,
author = { Mayo, Perla and Pirkl, Carolin M. and Achim, Alin and Menze, Bjoern and Golbabaee, Mohammad},
title = { { Physics informed guided diffusion for accelerated multi-parametric MRI reconstruction } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15975},
month = {September},
page = {432 -- 442}
}
Reviews
Review #1
- Please describe the contribution of the paper
The manuscript proposes an alternating optimization approach for solving the nonlinear inverse problem of recovering complex-valued images and tissue parameters from MRI measurements
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The theoretical formulation of the optimization is sound and well presented.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The biggest drawback is a lack of novelty in both training and inference using the diffusion. A prior work section is completely lacking.
Training is done using parameters for images (e.g. 1000 noise levels), without any justification or consideration for MRI signals.
The inference formulation in Eq. (5) is fundamentally the same as in e.g. Reference 20, except for the regularization term which involves the Bloch equations.
There are also no diffusion model baselines compared against from the ones cited in the paper.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
There is no mention of source code being released upon acceptance, and given the relatively large number of algorithm hyper-parameters and lack of justification for how they were chosen, this makes the paper very hard to reproduce.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Lack of novelty
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
The paper describes a diffusion-based model for reconstructing magnetic resonance fingerprinting (MRF) data with more accurate tissue parameter estimations and improved measurement fidelity. The contribution of the paper is to incorporate MR physics, 1) data fidleity of the acquired k-space and 2) Bloch equations, into the denoising diffusion model.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Detailed supporting theory of the proposed diffusion model.
- Novel apporach of combining diffusion model with MRF forward model.
- Comparisons between the propsoed approach and several existing methods.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- Limited applications of the apporach: 2D MRF with thick slice thickness acquired with 8-chan coil.
- Lack of rigorous experiments: 8 subjects with 15 axial slices were used for training and testing, where the subjects were not split during training and test.
- Too limited demonstration of the results: only one figure with single slice was demonstrated.
- Marginal performance improvements: The performance gain using the proposed model seemed to be marginal compared to other comparisons.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
- The pretrained diffusion model: The authors might need to clarify how the model was pretrained and it affects the overall performance in the end. Could the diffusion model be finetuned as well using the acquired MRF data? Since the noise characteristics and artifacts might be different between the data used for diffusion model pretraining and training/test using the acquired MRF data.
- Rigorous experiments are needed: subjects data need to be separated for training and test at least to fairly demonstrate the performance.
- Please provide more slices (from low to top of the brain coverage) to demonstrate the overall performance of the model.
- In Fig. 1, the performance difference between the MRF-DiPh and other methods (especially MRF-IDDPM or SCQ) seems to be very marginal. Demonstrating other subjects/slices might be needed to see the overall performance.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(3) Weak Reject — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Based on the points addressed above, Weak Reject is recommended for this paper.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The paper proposes MRF-DiPh, a novel reconstruction framework that integrates a pretrained denoising diffusion model with physics-based constraints for accelerated MRF. The key contribution lies in the combination of deep generative priors with explicit enforcement of k-space consistency and Bloch response fidelity within a proximal optimization framework. This hybrid approach is well-motivated, technically sound, and demonstrates strong performance improvements over existing methods on in-vivo brain data. The results suggest that MRF-DiPh effectively balances data fidelity and prior knowledge, making it a promising direction for quantitative MRI reconstruction under high acceleration.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
Novel formulation: The paper uniquely integrates a pretrained diffusion model with both k-space and Bloch model consistency, enabling physics-informed sampling—a first in MRF reconstruction.
Principled optimization: Uses a clear HQS-based proximal framework to alternate between learned priors and physics constraints, offering modularity and robustness.
Strong evaluation: Demonstrates consistent improvement over state-of-the-art methods across multiple metrics (T1/T2 MAPE, NRMSE) on in-vivo brain MRF data.
Clinical relevance: Inference is efficient (~44s), and the method is evaluated under realistic MRF settings, supporting practical feasibility.
Flexible conditioning strategy: Introduces a tunable mechanism to balance sampling diversity and reconstruction accuracy.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
Limited novelty in diffusion prior: Using pretrained diffusion models for inverse problems is known; the novelty mainly lies in its application to MRF. Missing strong baselines: Lacks comparison to model-based deep unrolled methods (e.g., MoDL-MRF).
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I recommend accepting this paper due to its novel and well-executed integration of physics constraints into a diffusion-based reconstruction framework for MRF. The method is technically sound, combining a pretrained denoising diffusion model with k-space and Bloch model consistency in a principled optimization scheme. This hybrid formulation is both elegant and practical, addressing key limitations of prior MRF approaches.
The experimental evaluation is strong, demonstrating clear improvements over existing baselines on in-vivo brain data, and the proposed method shows promising clinical feasibility with efficient inference time. While some components (e.g., diffusion priors) build on existing work, the application to quantitative MRI with dual physics constraints is new and impactful. Limitations such as 2D-only validation and lack of unrolled baseline comparisons are noted, but do not outweigh the contributions.
Overall, the paper makes a meaningful advance in MRF reconstruction and offers valuable insights into combining deep generative priors with physical modeling.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
Author addressed my questions.
Author Feedback
Novelty We thank the reviewers for their evaluations. R1 & R3 identified the novelty of our formulation and reconstruction approach as a major strength. R3 specifically highlighted the modular integration of diffusion models with MRF’s dual physics constraints via “principled optimization” as “new and impactful” and “elegant and practical, addressing key limitations of prior MRF methods”. While R2 raised concerns about novelty, we would like to clarify that our contribution does not lie in proposing new diffusion architectures or training schemes (we agree this area remains open for exploration), but in our inference framework that combines diffusion with MRF’s unique dual physics constraints. The algorithm in [20], so other diffusion-based CS-MRI methods, do not address the discretized Bloch constraints in Eq 5. A new algorithm was needed to fully handle these constraints, which we present as MRF-DiPh in §3.
Experiments R1 requested clarification on the training and data split. We apologize if this was unclear. §4 specifies a “75-25% train-test split,” which refers to mutually exclusive subjects—6 for training, 2 held-out for testing—ensuring no data leakage. R1 also asked about the acquired MRF data. Both train and test sets were acquired using the same scanner, acquisition/noise settings and MRF sequence [21] (§5 paragraph 1). The model was trained from scratch (no fine-tuning) using MRF training data; §3.1 (paragraph 2) details how (xc, xref) image pairs were obtained from MRF k-space acquisitions. At test time, only undersampled k-space y and its crude/aliased reconstruction xc = A^{H}y (both from test data) are used—y within proxf, and xc as side information input to the trained diffusion model. An unconditional model ignoring xc was also evaluated (ablation D, §5 paragraph 2, Tab1, Fig1) but performed worse.
R1 also noted paper’s comparisons as a key strength, as did R3 for evaluations under “realistic MRF settings” and against several MRF SOTAs, while noting the absence of an unrolled MRF baseline which we acknowledge. In response to R2’s concern about missing diffusion baselines, we clarify that we did compare to MRF-IDDPM [25], the only diffusion-based MRF method we found. Additionally, our ablations explored several key configurations: (1) MRF-IDDPM—without physics constraints; (2) MRF-DiPh (C)—with only k-space consistency, close to CS-MRI methods; and (3) MRF-DiPh (D)—an unconditional version without spatial guidance, aiming to highlight the role of diffusion under varying levels of physics/priors integration.
Results R1 pointed out Fig1 shows marginal gains of MRF-DiPh over SCQ & IDDPM. While most baselines perform reasonably on T1, T2 estimation is notably harder in short MRF sequences [21]. At 5x acceleration, differences are more evident in T2 maps (Fig1, Tab1). Zooming in Fig1 (recommended) shows MRF-IDDPM giving dark/underestimated T2 in white matter; SCQ lacks clear/sharp tissue boundaries on T2 in regions like pallidum, putamen, insula. Other baselines fail on T2, consistent with trends in Tab1. As discussed in §5, MRF-DiPh also yields notably better k-space fidelity over IDDPM—a critical advantage ensuring outputs adhere to actual measurements. We understand R1’s point on showing more slices, however, due to space and no-supplementary policy, we focused on one representative test case. If accepted, we will release source codes and demonstrator for additional reconstructions and will present these for further discussion at the conference.
Finally, validations used in-vivo 2D data to establish feasibility. 3D extension (raised by R1&R3) is indeed important and is an area of our ongoing investigation.
Once again, we thank the reviewers for their valuable feedback. We hope our responses have adequately addressed their concerns.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
This paper proposes a diffusion-based model for MRF reconstruction. While the proposed methods appear promising, the contribution of the paper is limited by the lack of extensive evaluation. Only two subjects were used for testing, with no cross-validation performed. Additionally, the quantitative results report only mean values without standard deviations, making it difficult to assess the true improvement of the proposed method compared to other methods.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A