List of Papers Browse by Subject Areas Author List
Abstract
Multi-Frequency Electrical Impedance Tomography (MF-EIT) is a non-invasive, low-cost modality that reconstructs electrical property distributions from boundary voltages. For stroke imaging, progress in 3D deep-learning reconstruction is limited by the lack of large-scale datasets with paired ground-truth (GT) volumes and by non-standardized pipelines for data generation, simulation, and evaluation. We introduce BREIT, a modular framework for 3D MF-EIT stroke reconstruction providing: (i) a neuroimaging-to-EIT pipeline that converts CT/MRI into frequency-dependent GT admittivity volumes; (ii) a self-contained Python 3D Complete Electrode Model (CEM) forward solver for simulating MF-EIT voltages; and (iii) a 3D D-bar implementation supporting non-uniform electrode layouts. Building on BREIT, we propose dFNO-bar, which integrates Fourier Neural Operators into D-bar by learning a mapping from scattering data $t(\xi)$ to conductivity $\sigma(x){=}\Re{\gamma}$. We evaluate dFNO-bar against D-bar, Deep D-bar, and Gauss–Newton reconstructions on UCLH-matched synthetic data, and observe higher brain SSIM with comparable CC across noise settings. Code and data are publicly available at: \url{https://github.com/djahiddj13/BREIT}
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2625_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/djahiddj13/BREIT
Link to the Dataset(s)
https://zenodo.org/records/20937062
BibTex
@InProceedings{AbdDja_BREIT_MICCAI2026,
author = { Abdelmoumene, Djahid AND Ayad, Ishak AND Nguyen, Maï K. AND Daveau, Christian},
title = { { BREIT: A Framework for Brain Stroke Reconstruction using Multi-frequency 3D EIT } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16895},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The main contribution of the paper is the introduction of BREIT for 3D multi-frequency electrical impedance tomography in brain stroke reconstruction. This work address a key limitations in data availability. Specifically, the work unifies the entire pipeline from CT/MRI-based generation of anatomically realistic, frequency-dependent ground-truth admittivity volumes, to accurate forward simulation using a 3D Complete Electrode 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.
The work addresses a critical gap in the field by introducing a fully integrated and reproducible framework. Unlike prior studies that focus on isolated components, this paper unifies data generation, forward simulation, reconstruction, and evaluation within a single pipeline. This improves reproducibility and enables fair comparison across methods, which has been a longstanding challenge in MF-EIT research.
- 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 has several weaknesses that limit its impact.
1) The evaluation is conducted primarily on synthetic data rather than real clinical measurements. Although the pipeline is grounded in CT/MRI and partially validated against real MF-EIT voltages, the reconstruction performance of dFNOBar is not demonstrated on actual patient data, which raises concerns about real-world generalization and clinical applicability.
2) I found figure 1 contains several missleading explanation expecially the use of arrows and inconsestencies in used MRI images.
3) The study depends heavily on simulated ground-truth admittivity maps and modeling assumptions (e. g. , tissue properties, segmentation accuracy, electrode modeling). Any inaccuracies in these assumptions may propagate through the pipeline and bias both training and evaluation, yet this limitation is not deeply analyzed. There are several studies the provide deep learning methods to estimate dielectric properties from MRI (like, DOI: 10.1109/TMI. 2020.2969682). How accurate is the estimation of tissue conductivity in comparison with SOTA methods?
4) While the framework is interesting, the novelty of the methodological contribution (dFNOBar) is somewhat unclear to me, as it builds on existing D-bar and Deep D-bar approaches by integrating Fourier Neural Operators, without fundamentally redefining the reconstruction paradigm. Explain this please.
5) The clinical validation and interpretability aspects are underdeveloped. There is limited discussion on how the reconstructed images would be used in clinical decision-making, how robust the method is to real-world noise/artifacts, or how uncertainty is handled.
6) Be sure to define all methematical notations used in equations.
7) Be sure to include link for open-source code ommitted in the abstract.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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.
(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?
The proposed method is validated mainly on synthetic data, with no convincing real clinical evaluation, raising concerns about practical applicability. The methodological novelty is limited, as it incrementally extends existing D-bar approaches without a clear breakthrough. Additionally, the experimental comparisons and reproducibility details are insufficient, weakening the overall scientific contribution.
- Reviewer confidence
Very confident (4)
- [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.
I am satisfied with author with author’s rebuttle.
Review #2
- Please describe the contribution of the paper
The paper’s main contribution is a enhanced reconstruction approach for stroke-related brain imaging using 3D multi-frequency EIT data. More specifically, it augments an available real 3D multi-frequency EIT dataset with a synthetic dataset generated from CT and MRI images and uses this dataset to develop and test a method that integrates Fourier Neural Operators into a d-bar approach for image reconstruction and the derivation of diagnostic findings related to brain stroke. A second important contribution is therefore not just the model itself, but also the dataset strategy: because real EIT datasets are limited, the work shows how synthetically generated data can be used as a valid and methodologically sound way to apply the approach for EIT 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.
Strong methodological combination: It combines a classical D-Bar approach with deep learning, specifically using Fourier Neural Operators together with D-Bar for image reconstruction. That gives the work both a solid model-based foundation and a modern learning-based component. Use of both real and synthetic data: The paper shows a way to deal with lack of real world EIT data. It uses real 3D multi-frequency EIT recordings as well as synthetic recordings generated from CT and MRI images, which strengthens the experimental setup. Transparent and adequate data-generation strategy: A further strength is that the paper explains how the synthetic data was generated in an appropriate way, which supports reproducibility and makes the approach more credible. Appropriate analytical and statistical evaluation: The benchmark selection resembles the state of the art. The analytical and statistical procedures were suitably chosen and correctly applied, which is a major strength for the validity of the results. Promising results with transfer potential: Even if the difference to related approaches is not large in every setting, the dFNObar reconstruction model would be worthwhile to transfer to other EIT application cases. That suggests broader potential beyond the specific study.
- 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.
Potentially modest performance gains over related methods: The evaluation suggests that the difference between the dFNObar-based approach and the DeepDBar approach is not very large, and is mainly visible for ischemia data. That means the empirical advantage of the proposed method may not yet be consistently strong across 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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.
(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?
My overall score was mainly driven by the combination of methodological soundness and practical relevance on the one hand, and left limited empirical strength on the other. The paper presents a technically interesting hybrid FNO/deep-learning approach for 3D multi-frequency EIT reconstruction and uses an reasonable synthetic-data strategy under realistic data constraints. The analytical and statistical procedures also are appropriate. Only the the empirical improvement over related approaches seems modest and most visible only in ischemia cases. Ideally the transferability and adaptability to other EIT application gets highlighted more. For that reason, I see the work as solid and promising.
- 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.
I still support the papers acceptance. The feedback to the rebuttal adresses the concerns of our reviews.
Review #3
- Please describe the contribution of the paper
This paper introduces BREIT, a modular framework designed for 3D multi-frequency EIT reconstruction in stroke imaging, along with a dedicated reconstruction model, dFNOBar. BREIT facilitates an end-to-end pipeline—from constructing frequency-dependent 3D admittance via CT/MRI to performing 3D CEM-based forward simulations and D-bar reconstructions for non-uniform electrode arrays. The dFNOBar model processes D-bar scattering data at multiple truncation radii, employing a hybrid architecture of 3D FNO and a lightweight 3D U-Net for high-fidelity conductivity mapping. Evaluations on synthetic datasets indicate that our approach outperforms state-of-the-art methods like D-bar, DeepDBar, and one-step Gauss–Newton in SSIM, while significantly optimizing parameter efficiency and memory footprint.
- 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 authors establish a standardized pipeline from neuroimaging to frequency-dependent admittance mapping, which effectively addresses the scarcity of high-quality paired datasets in EIT. The proposed architecture, which uses multi-truncation scattering data as input for an FNO, outperforms traditional Deep D-bar in SSIM (within synthetic environments) while maintaining a more compact parameter count. Furthermore, by constructing 3D datasets from public stroke data and external MRI/CT cohorts, the work moves beyond simplified 2D settings toward more clinically realistic scenarios.
- 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.
Heavy reliance on synthetic data: The core evaluation of the 3D reconstruction is conducted on self-generated synthetic data because public datasets lack electrode voltage measurements. Consequently, the method’s performance on real clinical data remains unproven.
Limited clinical validation: Clinical verification is currently restricted to the forward simulation level. Demonstrating consistency between simulated and measured voltages validates the simulation pipeline but does not guarantee that the inverse reconstruction will be effective in practice.
Marginal performance improvement: The performance boost of dFNOBar compared to DeepDBar is relatively small. There is a perceived gap between the complexity of the proposed innovation and the actual gains achieved.
Incomplete ablation and baseline analysis: The manuscript lacks a systematic ablation study to quantify the contributions of individual design elements (e.g., multi-truncation, FNO-U-Net architecture, electrode weighting). Furthermore, the fairness of comparisons with classical baselines should be more thoroughly discussed.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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 not only proposes a novel reconstruction model but also establishes a comprehensive framework for 3D MF-EIT stroke reconstruction and evaluation, demonstrating significant systematic depth and practical value. While the current results rely heavily on synthetic data and lack sufficient validation with real-world clinical datasets, the overall work is well-structured and the method shows consistent performance improvements.
- Reviewer confidence
Not confident (1)
- [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.
N/A
Author Feedback
We thank the reviewers for their constructive feedback and for recognizing the soundness and relevance of our framework. Due to page limits and the breadth of our contributions, some points could not be fully addressed in the initial manuscript.
[R1,R3] Evaluation on synthetic data. The public UCLH voltages are not directly compatible with our current complete-electrode D-Bar implementation, because usable drive-electrode voltages are not provided. Therefore, BREIT uses synthetic data matched to the UCLH clinical setup (electrode geometry, acquisition protocol, and CEM model) and grounded in real patient neuroimaging. We validated this by comparing simulated and measured clinical voltages, achieving mean r=0.82.While a synthetic-to-real gap remains, adapting the method to partial-boundary data is the natural route to direct D-Bar benchmarking on measured clinical data.
[R1,R3] Clinical validation and interpretability. Our approach is intended for complementary bedside/longitudinal monitoring rather than a replacement for CT/MRI. Interpretability is aided by using D-Bar scattering data as a physics-based intermediate representation rather than voltages directly. Robustness was evaluated under controlled noise levels comparable to the clinical voltage scale. Although a gap remains between forward validation and clinical inverse reconstruction, the simulated/measured voltage agreement supports the forward realism of the physical modeling used to generate the data. Uncertainty quantification and robustness to acquisition artifacts remain future work.
[R1] Simulated and modeling assumptions. We agree that the image-derived admittivity pipeline is approximate, and that modeling choices can affect the generated reference maps and simulated voltages. BREIT keeps these choices explicit and modular. The method cited by R1 is better viewed as a learned surrogate for segmentation-to-conductivity mapping, rather than an independent ground-truth estimator. It is trained from segmentation-derived conductivity targets, and its smoothing is not established to be closer to true physical conductivity. Our pipeline instead keeps the assumptions transparent and tunable and tests their forward consequences against measured MF-EIT voltages.
[R1] Methodological novelty. The novelty is changing the information available to the learned model. D-Bar stabilizes reconstruction by truncating high-ξ scattering components, which are more noise-sensitive but may also contain useful structural information. In contrast, dFNOBar provides several complex scattering truncations Tξ as input channels, allowing the network to learn which higher-frequency components to retain, suppress, or denoise. The FNO choice was motivated by its suitability for learning PDE-related solution operators, matching the final D-Bar recovery stage that maps scattering-domain information to conductivity.
[R2,R3] Empirical improvements. In Fig. 5, dFNOBar consistently outperforms DeepDBar across all pathology groups, with the largest gains in ischemic stroke. SSIM improvements are statistically significant across noise levels, with average SSIM increase of +3%. dFNOBar is also more efficient, using 25% fewer parameters and 14% less memory.
[R3] Additional ablation. We agree that component-wise ablations would strengthen the analysis. The submitted evaluation focuses on standardized end-to-end comparison under the same BREIT data, grid, and train/test protocol. The design choices follow the method: multi-truncation uses discarded scattering information, FNO learns the scattering-to-conductivity map, U-Net refines artifacts, and electrode weighting handles non-uniform electrode layouts.
Minor revisions. We will revise Fig. 1 and MRI illustrations, define all symbols, add the code link to the abstract, and clarify admittivity pipeline assumptions.
We hope BREIT will support reproducible MF-EIT stroke research and development of more robust deep-learning reconstruction models.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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.
Strengths: The paper offers an integrated and reproducible 3D multi-frequency EIT stroke reconstruction pipeline, combining model-based D-Bar ideas with Fourier Neural Operators and a transparent synthetic-data generation strategy informed by real neuroimaging and EIT measurements. Weaknesses: Its impact is limited by heavy reliance on synthetic reconstruction evaluation, limited real clinical validation, only modest gains over DeepDBar, unclear methodological novelty in parts, incomplete ablation/baseline analysis, and some presentation or notation issues. The authors therefore be invited for the rebuttal.
- 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.
All reviewers reached to a consensus to accept after the rebuttal.
Meta-review #2
- 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.
The multi-frequency 3D EIT reconstruction framework for brain stroke addresses a clinically urgent problem, and the validation across phantom and in silico scenarios provides a solid methodological foundation.
The rebuttal achieved near-consensus from a previously split profile supporting the paper’s underlying quality and the authors’ outstanding rebuttal. I would to emphasize the rebuttal’s quality by the turnover of R1, from reject to accept, but also R3 upgraded to accept.
Meta-review #3
- 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.
All reviewers are satisfied with the rebuttal and agree that this paper has good merit and recommend acceptance. It is a clear accept.
Meta-review #4
- 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.
This paper proposes a very interesting method for a fully integrated pipeline in 3D stroke reconstruction. All the concerns notably related to the use of synthetic data, the performance and the methodological novelty are very well addressed in a highly commendable rebuttal