List of Papers Browse by Subject Areas Author List
Abstract
Low-field (LF) magnetic resonance imaging (MRI) improves accessibility and reduces costs but generally has lower signal-to-noise ratios and degraded contrast compared to high field (HF) MRI, limiting its clinical utility. Simulating LF MRI from HF MRI enables virtually evaluating novel imaging devices and developing LF algorithms. Existing low field simulators rely on noise injections and smoothing, which fails to capture the contrast degradation seen in LF acquisitions. To this end, we introduce an end-to-end LF-MRI synthesis framework that learns HF to LF image degradation directly from a small number of paired HF-LF MRIs. Specifically, we introduce a novel \textbf{H}F to \textbf{L}F coordinate–image decoupled neural \textbf{O}perator (H2LO) to model the underlying degradation process, and tailor it to capture high-frequency noise textures and dedicated image structure. Experimental results in T1- and T2-weighted MRI demonstrate that H2LO produces more faithful simulated low-field images than existing parameterized noise synthesis model and popular image-to-image translation models. Furthermore, it improves performance in downstream image enhancement tasks, showcasing its potential to enhance LF MRI diagnostic capabilities.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0707_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{GaoZiq_SubjectSpecific_MICCAI2026,
author = { Gao, Ziqi AND Dvornek, Nicha C. AND Zhang, Xiaoran AND Galiana, Gigi AND Tagare, Hemant D. AND Constable, R. Todd},
title = { { Subject-Specific Low-Field MRI Synthesis via a Neural Operator } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16890},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
-
Authors propose a model that synthesizes ULF images from HF images using a neural‑operator–based network that models the underlying degradation process and captures high‑frequency noise textures and relevant image structure. Their approach employs a branch–trunk architecture optimized with pointwise intensity fidelity and gradient‑based regularization.
-
They also evaluate a downstream image‑enhancement task, using the synthesized images as augmentation to assess whether the method improves performance on real ULF data.
-
- 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 application of neural operators to image synthesis is a key novelty of this work.
The authors claim that their model addresses limitations of prior pixel‑wise image‑to‑image translation methods, which often fail to capture realistic textures and accurately model contrast changes. By explicitly modeling the degradation process, the proposed approach provides a more principled way to approximate ULF acquisition characteristics.
- 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.
While the proposed approach introduces a novel use of neural operators for ULF synthesis, the empirical improvements over existing state‑of‑the‑art methods are relatively marginal. This limits the current impact of the work and suggests that the method requires further refinement to demonstrate clear advantages in real‑world settings.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/en/REVIEWER-GUIDELINES.html
Figure 2 is referenced in the Introduction, but the figure itself is primarily presented in the Results section. In addition, the authors could include a dedicated discussion section to address the limitations of current state‑of‑the‑art methods in comparison to their approach. The manuscript would benefit from reorganizing these elements to ensure a more consistent and coherent reading experience.
Why do you compare only the results of Arnold’s work in Figure. 2? Is it due to lowest parameters (=4) used in the model?
- 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?
Given the strengths and weaknesses outlined above, I believe the manuscript should be accepted for a poster presentation. Authors provided comprehensive experimental evaluations, and a poster format will allow them to present their ideas, receive targeted feedback, and further refine this new direction for the field.
- 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 #2
- Please describe the contribution of the paper
The paper proposes a neural operator-based framework (H2LO) for subject-specific low-field MRI synthesis, reformulating HF-to-LF translation as a continuous operator learning problem. By combining a coordinate-aware neural operator with high-frequency-preserving design, the method captures both contrast shifts and noise characteristics more effectively than conventional simulators and image-to-image models, while also demonstrating utility for downstream enhancement tasks. Personally I appreciate the idea, from HF to LF, which I believe is under-explored and valuable.
- 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.
First, the paper presents a novel and well-motivated formulation by modeling HF-to-LF MRI synthesis as a continuous operator learning problem, rather than a conventional voxel-wise image translation task. This perspective is aligned with MRI physics, where field strength affects the underlying signal formation and frequency characteristics, making the formulation both principled and meaningful. Second, the proposed H2LO architecture is technically elegant and well-designed, combining a CNN-based branch network with a coordinate-based SIREN trunk to capture both spatial context and high-frequency details. The introduction of spatially varying operator coefficients allows the model to adapt locally to tissue-specific contrast changes, which is particularly important for realistic LF simulation. Third, the paper demonstrates practical relevance and downstream utility, showing that the synthesized LF images can effectively augment training data and improve LF MRI enhancement performance. The evaluation is relatively comprehensive, including both reconstruction metrics and histogram-based measures, and shows consistent improvements over traditional simulators and image-to-image baselines
- 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.
First, while the operator-based formulation is well motivated, the methodological novelty over INR and operator-learning works is somewhat limited. The proposed H2LO builds upon existing frameworks such as DeepONet and coordinate-based representations (e.g., SIREN, LIIF-style extensions ), and the contribution mainly lies in adapting these components to the HF to LF MRI setting rather than introducing a fundamentally new architecture.
Second, the study relies on a very limited dataset (23 subjects) for learning the HF-to-LF mapping, which raises concerns about generalization and robustness. Although 5-fold cross-validation is used, the model may still overfit to scanner-specific characteristics or acquisition protocols, and the paper does not provide evaluation on external datasets or across different scanners/field strengths .
Third, the baseline comparison, while reasonable, is not fully comprehensive, particularly with respect to recent medical imaging methods. The evaluation focuses on general I2I models (Pix2Pix, diffusion) and a classical simulator, but lacks comparison with more recent physics-informed or diffusion-based MRI translation/enhancement approaches (e.g., works such as ref 8,9,11 cited in the paper but not included as baselines), which may limit the strength of the empirical claims.
Fourth, the paper does not fully establish clinical validity or realism of the synthesized LF images, beyond standard image similarity and histogram-based metrics. There is no expert evaluation or task-based assessment (e.g., diagnostic performance), making it unclear whether the generated images faithfully reflect clinically relevant LF characteristics rather than metric-level similarity.
Finally, the method assumes perfect spatial alignment between HF and LF scans, which may not hold in real-world scenarios. The reliance on registered paired data simplifies the problem and may limit applicability to settings where such precise alignment is unavailable or difficult to obtain.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper presents a technically sound and well-motivated approach to HF to LF MRI synthesis, with a formulation that is aligned with MRI physics and a clean neural operator design. The idea of modeling cross-field transformation as an operator is both conceptually meaningful and practically useful, especially given its demonstrated benefit for downstream tasks like LF enhancement.
However, the contribution is somewhat incremental in methodology, as it builds on existing neural operator and implicit representation frameworks, and the evaluation is limited by a small dataset and lack of clinical validation. These issues prevent it from being a strong accept, but the overall clarity, technical consistency, and relevance make it worthy of acceptance.
- 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 an end-to-end framework for synthesizing low-field (LF) MRI from high-field (HF) MRI by formulating the problem as a mapping between function spaces rather than fixed-grid voxel regression.
- Specifically, the authors model the HF-to-LF transformation as a neural operator. The proposed architecture consists of a 3D CNN-based branch network that encodes the input HF volume, and a sinusoidal implicit neural representation (SIREN) trunk network that takes output spatial coordinates as input. The outputs of the branch and trunk networks are combined to predict the LF image intensity at a given location.
- The method demonstrates improved performance over conventional image-to-image translation approaches while using fewer parameters.
- 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.
- Novelty: This is the first paper proposing an end-to-end framework for HF-to-LF MRI image synthesis
- Performance: The method generally outperforms all baselines both quantitatively and qualitatively, while using the least number of parameters.
- Elegant: The formulation of HF-to-LF synthesis as a function to function mapping constraints the model to mathematically backed good solutions compared to pixel-to-pixel 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.
- Requires paired HF-LF images to train which can be challenging to obtain in practice.
- It is unclear whether the proposed method can generalize to unseen HF-to-LF scenarios i.e. if a model is trained on HF dataset A, can it equally synthesis LF images from a new dataset B. Dataset B could be from a different anatomy or device.
- The proposed SIREN trunk is inferior to the w/o SIREN setting based on the quantitative results and ablations, and its justification is based on subjective qualitative analysis. Also, it is not clear what w/o SIREN setting exactly is.
- The method was only evaluated on a single dataset of healthy subjects and it remains to be seen how the method generalizes to large popultations, different devices, sites, and pathological cases.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/en/REVIEWER-GUIDELINES.html
Minor comments:
-
Histogram metrics: the metrics are computed on the foreground (IHF > 0.01). This could potentially explain why the metrics show Pix2Pix to be better than the proposed method as T2W images are mostly dark. Perhaps, for future work, you want to modify the definition of the background to mean the entire interior of the skull.
-
“…,our method achieves the best performance across almost all metrics for both T1 and T2 contrasts”: The method achieves the best results across all metrics only on T1.Please correct.
-
What does the arrow from the branch output to the trunk input represent? My understanding is that the trunk only expects x as the input.
-
- 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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed method is novel and adopts DeepONet to a new medical problem of low-field MRI image synthesis, a largely underexplored domain.
- 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
Author Feedback
We thank all reviewers and the meta-reviewer for the efforts in reviewing this paper and providing constructive and positive feedback. We are very encouraged by the fact that reviewers found both the methodology, neural operator, and the application venue, LF MRI synthesis, reasonable and interesting.
We would like to answer some of the reviewers’ points of confusion.
R1 12, paragraph 2: In histogram visualization, we compare against Arnold et al.’s method because their model is explicitly optimized for histogram-statistic matching. In contrast, our model and the other comparison models are trained with pixel-wise consistency objectives. Therefore, we considered it more appropriate to visualize Arnold et al.’s method using histogram-level consistency, which better aligns with the objective it optimizes. We will clarify this point in the camera-ready version. R3 12, paragraph 3: We apologize for drawing the arrow in the wrong direction and causing confusion. Yes, the trunk only expects coordinates, and the coordinate lookup is the information flow from the trunk to the branch. We will correct this in the camera-ready version.
For the rest of the constructive feedback, we will incorporate it into our future work to further improve the method and clinical utility. Thanks again for your kind reviews.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
Given that all three reviewers have recommended acceptance, I am pleased to congratulate the authors on the acceptance of their paper at this phase.