List of Papers Browse by Subject Areas Author List
Abstract
Magnetic Resonance Elastography (MRE) is a non-invasive imaging modality quantifying soft tissue stiffness. The reconstruction of stiffness maps is based on solutions of an inverse problem, which poses challenges in balancing accuracy, computational resources, and robustness. To stabilize the reconstruction, many inversion techniques, and most recently neural network-based inversion techniques, have explored multifrequency acquisition and reconstruction. However, these techniques typically perform separate single-frequency inversions followed by multifrequency aggregation. In this work, we propose a fully multifrequency neural network-based inversion trained on synthetically generated data that directly incorporates the relationship between multifrequency acquisitions, assuming a viscoelastic material model. Our proposed approach provides flexibility with respect to the acquisition frequencies, ensuring its practical applicability in the clinical and research setting. We evaluated our method using finite element simulations and in vivo abdominal MRE datasets, achieving increased accuracy and providing a more reliable and effective solution for MRE-based tissue characterization than standard reconstruction approaches.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3156_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)
All datasets: https://bioqic-apps.charite.de/
BibTex
@InProceedings{BusHél_Multifrequency_MICCAI2025,
author = { Bustin, Héloïse and Meyer, Tom and Jordan, Jakob and Walczak, Lars and Tzschätzsch, Heiko and Sack, Ingolf and Hennemuth, Anja},
title = { { Multifrequency Neural Network-based Wave Inversion in MR Elastography } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15962},
month = {September},
page = {422 -- 432}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose an original approach for Magnetic Resonance Elastography, based on neural networks to reconstruct shear wave speed maps (shear wave inversion, necessary for elasticity mapping), accounting for multi frequency acquisition.
- 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 manuscript is very well presented and documented, and the method rigorously so. This kind of approach is entirely relevant, and the comparison with more usual methods is properly documented and carried out.
- 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.
It might be useful to have a little more detail on the simulations used to create the neural network training database. In particular, how source excitation and, even more so, boundary conditions are taken into account (interactions and wave reflections at organ boundaries, anchor points, etc.). The question also arises when extrapolating to organs other than the liver or kidneys.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 link is very well provided, but not anonymously.
- 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?
The manuscript is very well presented and documented, and the method rigorously so. This kind of approach is entirely relevant, and the comparison with more usual methods is properly documented and carried out.
- 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 proposes a new Magnetic Resonance Elastography (MRE)-based sheer wave speed reconstruction method, based on a deep neural network. The paper proposes to combine multiple frequencies and components of the acquired wave displacement data directly into the network architectural design to model the inversion process (wave displacement -> sheer wavelength) accurately.
- 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 paper is very well written, easy to follow, and covers a broad variety of literature review and logically builds upon its single frequency counterpart, ElastoNet.
-
The idea of directly combining multiple frequency and multicomponent data into the architectural design of an attention-based CNN is interesting and shows that with increasing the data complexity of the network, along with a suitable fusion from the network architectural blocks (attention in this case) can lead to improved reconstruction of the sheer wave speed.
-
The paper is very clear in its formulation and the presented results thoroughly showcase the efficacy of the proposed method on a variety of simulated and real in-vivo datasets.
-
The flexibility of supporting multiple frequencies and range (via a ratio based power-law constraint), I believe, could be useful for clinical applications of MR across different lab and equipment settings.
-
- 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 paper did not exactly specify how many (u, lambda_s) pairs were used in the training of MF-ElastoNet. Also, it is not exactly clear how the training data differs from the FEM based testing datasets.
-
The ablation study is only conducted on the multiple and single frequency aspect; however the network architecture includes attention at three levels: spatial, component and frequency and it is not clear how the different attention blocks influence the performance globally.
-
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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
-
I feel that the authors could comment on the usefulness of values in Table 2. Though the ground-truth data is not available for the in-vivo data, but it would be useful to discuss the relevance of the results, as compared to standard sheer wave speed values for the different ROIs included in the in-vivo datasets.
-
In Table 3, I think that the authors should explain how the single-frequency version of the network was trained and how many training dataset points exactly were both versions trained for. Especially if the former was trained on only single frequency data, did the authors accordingly train it for more epochs/training data for a fairer comparison?
-
In Fig. 2, it is not immediately clear what n, d, h represent. Similarly, the sizes of the intermediate outputs of the layers/blocks could be shown to enhance clarity. Do the authors mean 500000 epochs in Section 3.1 instead? Also, I feel that the authors could report the reconstruction and training time of their method, in comparison to the other baseline methods.
-
- 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?
The paper presents a useful MRE-based sheer wave speed estimation method, based on a novel deep learning architecture. The network proposed fuses the multifrequency and multicomponent wavefield data directly into its structure, leading to enhanced speed reconstruction, as validated on multiple in-vivo datasets. The additional advantage of supporting multiple frequencies could be useful for clinical use for supporting different MR equipments.
- 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 #3
- Please describe the contribution of the paper
The paper introduces MF-ElastoNet, a fully multifrequency neural network-based inversion framework for magnetic resonance elastography. Unlike existing approaches that process each frequency separately and then aggregate the results, MF-ElastoNet directly integrates multifrequency data during training and inference. The authors model viscoelastic material behavior and use a hierarchical vision transformer architecture to reconstruct shear wave speed (SWS) maps from complex wavefields. The method is evaluated on both simulated and in vivo datasets, demonstrating improved accuracy, stability, and anatomical detail compared to both classical inversion techniques (LFE, k-MDEV) and recent deep learning approaches (ElastoNet, TWENN).
- 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.
1) Novel formulation: The work is the first to present a fully multifrequency neural network-based wave inversion approach in MRE, addressing a long-standing limitation in patch-based learning from single-frequency data. 2) Flexible frequency handling: The architecture accommodates variable numbers and values of frequencies, enhancing adaptability across clinical and research protocols. 3) Physics-informed synthetic training: The use of viscoelastic material modeling and constrained synthetic data generation based on power-law tissue behavior enhances realism and generalizability. 4) Strong evaluation: The method is comprehensively validated across FEM simulations and in vivo datasets, showing quantitative improvements in RMSE and visual gains in anatomical accuracy over strong baselines. 5) Transformer architecture: The use of hierarchical attention blocks for spatial, displacement component, and frequency-level fusion is elegant and well-motivated.
- 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.
1) Limited interpretability: Although accuracy improves, the model remains a black box. Interpretability methods or uncertainty quantification, as explored in other works (e.g., ElastoNet), would add value. 2) Comparison fairness: The comparison to other deep learning models (e.g., ElastoNet and TWENN) might be affected by differences in training setup (e.g., use of multifrequency vs. single-frequency training), and this should be clarified in the discussion. 3) Ablation analysis could be extended: The ablation study shows the benefit of multifrequency training, but further analysis of architecture components (e.g., transformer layers, attention blocks) would help isolate contributions.
- 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?
This paper presents a novel and well-executed approach to a relevant problem in MRE, demonstrating both conceptual and empirical innovation. The integration of multifrequency information within a transformer-based neural network framework is timely and effective. The method is thoroughly evaluated and outperforms prior work across multiple datasets. While some clinical validation and interpretability aspects could be improved, the overall contribution is significant and highly relevant to the MICCAI community.
- 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
Author Feedback
We thank the reviewers for their insightful and valuable feedback. We are pleased that they acknowledged the novelty and clinical relevance of our work. We address their concerns as follows: R1.1: Thank you for your comment regarding the training data generation. While our current data generation model did not incorporate potential reflections at elastic interfaces within a patch, this would be interesting to explore in future work. R1.2: Thank you for your comment. We would like to clarify that the link provided is to an open-access platform used by the MRE research community, where working groups can share their methods and data. R2.1: Thank you for your comment regarding the training data. The total number of generated training samples corresponds to the product of the batch size and the number of epochs. We will include this information in the methods. R2.2 and R3.3: Thank you for your comments regarding the ablation study. Due to limited space in the paper, we chose to focus the ablation study on the multiple vs single frequency aspect of the method, which is a complete novelty in the field of neural network-based MRE inversions. R2.3: Thank you for your comment regarding the relevance of Table 2. While there was no ground truth available for in vivo datasets, in the context of future translation to clinical use, we found it essential to illustrate how our method performs on in vivo datasets. The regions of interest show the ability of the different methods to resolve regional differences.
R2.4: Thank you for your question regarding Table 3. The training parameters were kept unchanged, and the single-frequency model was trained for the same number of epochs as the multifrequency network. This should actually disadvantage the multifrequency model, which was shown one tenth of the data per number of frequencies in comparison to the single-frequency network. R2.5: Thank you for pointing out the missing legend as well as the notation error in the number of epochs. We will correct the text in the final version accordingly. R3.1: Thank you for your comment regarding the limited interpretability of the model. We would indeed like to explore this in future work. We will add it to the discussion and conclusion. R3.2: Thank you for your suggestion regarding the comparison fairness. The neural network-based approaches in MRE are indeed influenced by an array of factors, including differences in the training setup. This complexity must be noted when considering the performance of these methods. We will clarify this information.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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