Paper Info Reviews Author Feedback Meta-Review Back to top

List of Papers Browse by Subject Areas Author List

Abstract

Extreme limited-view photoacoustic tomography reconstruction suffers from severely missing features and dominant strong artifacts, which often leads to structural discontinuities, blurred boundaries, and loss of fine details. Moreover, channel availability can vary in real acquisitions, making methods that rely on early multi-channel concatenation and fixed input configurations less stable. To this end, we propose S²-Mamba, which decouples the multi-channel input into multiple single-channel branches to preserve weak but informative cues within each channel. This design naturally accommodates missing channels and changes in acquisition configurations. Furthermore, we introduce a Cross-Channel Constrained Loss that explicitly aligns multi-channel representations at the latent distribution level and mitigates representation collapse through a uniform prior constraint. We also propose an Adaptive Gradient-Gated Loss that uses progressive hard mining to emphasize structurally relevant difficult regions and enables joint optimization via uncertainty-adaptive weighting. Under an extreme limited-view setting on an in vivo mice dataset, we train the model using 4-channel (11.25°) observations. The results show that our method outperforms a range of representative networks in terms of SSIM, PSNR, FID, and KID, and remains superior even when the channel configuration differs between training and testing, demonstrating its robustness and practical deployability.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/6667_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/USTCXSC/C2-Mamba-PAT

Link to the Dataset(s)

In vivo mice dataset: https://github.com/USTCXSC/C2-Mamba-PAT

BibTex

@InProceedings{XuShi_Crosschannel_MICCAI2026,
        author = { Xu, Shicheng AND Li, Jincheng AND Lan, Hengrong AND Gao, Fei},
        title = { { Cross-channel Consistent Mamba for Extreme Limited-View Photoacoustic Tomography Reconstruction } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16888},
        month = {September},
        page = {pending}
}

Reviews

Review #1

Please describe the contribution of the paper

The paper proposes S²-Mamba, a reconstruction framework for extreme limited-view photoacoustic tomography that explicitly addresses varying and missing channel configurations. The method decouples multi-channel inputs into shared single-channel branches, introduces a cross-channel consistency loss to align latent representations, and designs an adaptive gradient-gated loss to emphasize structurally difficult regions during training, aiming to improve robustness and structural fidelity under severely undersampled conditions.
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. The paper tackles a challenging and practically relevant problem, namely extreme limited-view reconstruction with missing or varying channel configurations, which is often overlooked in existing work.
2. The shared single-channel design is well motivated and provides a simple yet effective way to preserve weak per-channel information while naturally handling channel mismatch at inference time.
3. The overall framework is coherent, with architectural design and loss functions consistently targeting robustness and structural recovery under severe information loss.
4. The experimental setup is reasonably convincing, especially the extreme limited-view setting and the channel-mismatch evaluation, which highlights the practical advantage of the method.
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. The methodological novelty is somewhat limited. Most components, including the shared encoder design, cross-channel alignment, and hard-mining-based loss, are based on existing ideas, and the contribution mainly lies in their combination for this specific problem.
2. The proposed losses (cross-channel consistency and gradient-gated loss) are reasonable but not fundamentally new, and their individual contributions beyond standard ablation remain somewhat incremental.
3. The evaluation is conducted on a relatively small in vivo dataset, which may limit the assessment of generalization to more diverse or realistic clinical scenarios.
4. The method is somewhat tailored to the specific PAT setting, and its applicability to other reconstruction problems is not discussed.
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.
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?

The paper addresses a challenging and practically relevant problem in extreme limited-view photoacoustic tomography and proposes a coherent framework that explicitly handles missing and varying channel configurations. The method is technically sound, and the experimental results demonstrate consistent improvements, particularly in challenging settings such as channel mismatch. However, the overall methodological novelty is limited, as the approach mainly combines existing architectural and loss design ideas. In addition, the evaluation is relatively limited in scale. Overall, the work is well motivated and practically useful, but falls slightly short of a stronger acceptance due to its incremental nature.
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 authors propose the s2-mamba model with cross-channel constrained loss and adaptive gradient-gated loss for limited-view PAT reconstruction. The main contribution lies in the proposed optimization strategy, which adaptively suppresses gradients shifting during training. Both qualitative and quantitative results demonstrate the effectiveness in recovering fine microstructural details.
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 proposed adaptive gradient-gated loss incorporates the forward imaging model to mitigate over-smoothing and improve reconstruction accuracy. The introduction of the mamba module and channel interaction operations is well-suited to the limited-view learning objective. Overall, the method is technically sound and achieves strong performance.
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.The authors should include a color bar indicating pixel intensity value in Fig. 1, and add scale bar to Fig. 4.2.It is recommended to use distinct colors to differentiate abstract module connections from the detailed process flows shown above in Fig. 2.3.It is unclear whether Eq. (1) assumes n_i ≤ n_j or n_i < n_j. if u_c is defined as a vector, then its dot product with p_c in Eq. (2) appears inconsistent. The authors should verify the correctness and consistency of definitions in equations. 4.The evaluation metrics should be properly cited. The comparative models used in the experiments should be cited at their first mention. 5.In Fig. 4, it is unclear why the FWHM values for UKAN and the proposed method are higher than those of the ground truth. The authors should provide an explanation for this.
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.
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?

The proposed reconstruction pipeline is novel and effective, with technically sound and impactful modules and optimization strategies. Experiments on in vivo data demonstrate significant performance improvements, and the ablation studies are well-designed. The method is practically promising, with clear potential for advanced PAT imaging. The manuscript would benefit from minor revisions to improve clarity and technical detail.
Reviewer confidence

Somewhat confident (2)
[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

This paper proposes S²-Mamba for extreme limited-view PAT reconstruction. The method separates multi-channel back-projected inputs into shared single-channel branches, uses a Mamba-based encoder-decoder backbone, and introduces cross-channel consistency and adaptive gradient-gated losses to improve reconstruction robustness under missing or mismatched channel configurations.
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 addresses an important and practically relevant problem in limited-view PAT reconstruction. Extreme limited-view acquisition can cause severe streak artifacts, structural discontinuities, and loss of fine vascular details, and the paper is well motivated in this regard. A strength of the proposed framework is its channel-wise design. Instead of directly concatenating all channels at the input stage, the method processes each channel separately with shared encoder weights and fuses features at multiple scales. This design is reasonable for handling weak but complementary cues from different angular subsets and may improve robustness when channels are missing or acquisition configurations vary. The cross-channel consistency loss is also a meaningful component. It explicitly encourages latent representations from different channels to be aligned while using a uniform-prior regularization to reduce representation collapse. The adaptive gradient-gated loss is another useful idea, as it focuses learning on structurally difficult regions rather than allowing large smooth background regions to dominate the optimization. These losses appear particularly relevant for artifact-dominated limited-view PAT reconstruction.
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 motivation for using Mamba as the core backbone is not sufficiently convincing. Although Mamba may provide efficient long-range spatial modeling, the paper does not clearly demonstrate that Mamba is essential for the proposed reconstruction task. The main improvement appears to come more from the shared single-channel design and the proposed losses than from the Mamba backbone itself. The ablation study does not sufficiently isolate the contribution of Mamba. In particular, the performance of the backbone alone does not appear clearly superior to other representative reconstruction networks. This raises the question of whether Mamba is truly the key technical factor or whether similar gains could be achieved using a CNN- or Transformer-based backbone with the same channel-wise architecture and losses. The proposed losses are not specific to Mamba. The Cross-Channel Consistency Loss can be applied to any multi-branch network that extracts channel-wise latent features, and the Adaptive Gradient-Gated Loss can be used with any image-domain reconstruction framework. Therefore, the paper should better disentangle the respective contributions of the Mamba backbone, the shared single-channel architecture, and the proposed loss functions. Another significant limitation is the dataset size and subject diversity. Although the training and test animals are independent, the test set comes from only one mouse. This makes it difficult to assess generalization across biological variability, imaging conditions, and anatomical differences.
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?

I recommend weak accept. The paper addresses an important problem and proposes a reasonable channel-wise reconstruction framework with useful loss functions. However, the central motivation for using Mamba is not sufficiently validated, and the current experiments do not clearly separate the contribution of Mamba from the proposed channel-wise architecture and losses. In addition, the evaluation is limited by the very small number of test subjects, especially the use of only one mouse for testing.
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 sincerely thank the meta-reviewer and all reviewers for their positive evaluation and constructive suggestions. We are grateful that the reviewers recognized the practical importance of extreme limited-view PAT reconstruction, especially under missing or varying channel configurations, as well as the consistent quantitative and qualitative improvements of our framework. We have carefully revised the manuscript to improve clarity, presentation, and positioning of the contributions: 1.Regarding the concerns about methodological novelty, we agree that several components, such as shared feature extraction, cross-channel representation alignment, and hard-region mining, are inspired by existing ideas. Our intention is not to claim each component as fundamentally new in isolation, but to integrate them into a coherent framework specifically designed for artifact-dominated extreme limited-view PAT reconstruction. We have revised the Introduction and Conclusion to better position the contribution as a task-oriented framework that combines channel-wise reconstruction, cross-channel consistency, and adaptive structural optimization for this challenging setting. 2.Regarding the motivation and contribution of Mamba, we agree that the current experiments do not fully isolate Mamba from the overall channel-wise architecture and loss design. In this work, Mamba is adopted as an efficient long-range modeling backbone because vascular structures in PAT require global continuity and contextual modeling. However, we acknowledge that the proposed cross-channel consistency loss and adaptive gradient-gated loss are not restricted to Mamba and may also benefit CNN- or Transformer-based multi-branch reconstruction networks. Due to the limited scope of the camera-ready revision, we have not added extensive backbone-replacement experiments, but we have clarified this point in the revised manuscript and will further investigate this comparison in future work. 3.Regarding the dataset size and subject diversity, we acknowledge this limitation. Compared with mature imaging modalities, large-scale public PAT datasets remain relatively limited, and in vivo small-animal datasets are still more accessible than human clinical datasets. Although the training and testing animals are independent, the test set still comes from one mouse, which may limit the evaluation of generalization across biological variability, anatomical differences, and imaging conditions. We have added this limitation and future work discussion, and will further extend the validation to more subjects and, when available, human PAT data. 4.Regarding the applicability to other reconstruction problems, our current motivation is focused on the extreme limited-view PAT scenario, where severe angular information loss and channel mismatch are the central challenges. We have therefore not validated the method on other imaging modalities in this paper. We have revised the discussion to clarify that extending the proposed channel-wise framework and loss design to other reconstruction tasks is a promising direction for future work. 5.We also thank R#2 for the detailed presentation and notation suggestions. We have revised the manuscript based on the suggestions. For the mathematical notation, we have revised the notation in Eq. (2) to avoid ambiguity caused by the previous use of n_i in Eq. (1). Finally, we revised the description of Fig. 4 to avoid ambiguity regarding FWHM values. The higher FWHM values observed for UKAN and our method should not be interpreted as exceeding the resolution of the 512-channel ground truth, but rather as profile broadening caused by learned structural smoothing and boundary recovery under severe limited-view artifacts. The corresponding text has been modified to make this interpretation clearer.

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.

The reviewers agree that the paper addresses a practically important and challenging problem in extreme limited-view photoacoustic tomography, particularly under varying and missing channel configurations. The proposed framework is technically sound and demonstrates consistent improvements in both quantitative and qualitative evaluations, especially under channel-mismatch settings. While concerns were raised regarding the level of methodological novelty and the limited scale of evaluation, these issues are viewed as incremental rather than fundamental. Overall, given the practical relevance, coherent design, and solid empirical performance, the paper meets the bar for acceptance, pending minor revisions to improve clarity and better position the contributions.

back to top

Cross-channel Consistent Mamba for Extreme Limited-View Photoacoustic Tomography Reconstruction

Author(s):