List of Papers Browse by Subject Areas Author List
Abstract
Longitudinal glioma registration faces a fundamental Correspondence - Regularity Conflict: healthy parenchyma requires high stiffness to prevent folding, while tumor evolution demands local plasticity to accommodate mass effect. Current deep learning frameworks, treating the brain as a mechanically homogeneous continuum, fail to resolve this paradox. We propose PathoMamba, a stiffness-modulated State Space Model (SSM) that bridges continuum mechanics and deep learning. By reinterpreting the SSM discretization step (∆) as learnable Numerical Inertia, we actively modulate feature dynamics—enforcing rigidity in healthy tissue while enabling rapid updates in the pathology. This is coupled with a Therapy-Aware Biomechanical objective to robustly model heterogeneous growth and shrinkage trends. Validated on the BraTS-Reg benchmark, PathoMamba matches state-of-the-art Transformer accuracy (1.88 mm) while inferring 2× faster and uniquely guaranteeing 0.00% topological folding, offering a safe, physics-informed solution for clinical monitoring.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5198_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)
BibTex
@InProceedings{AsiErn_PathoMamba_MICCAI2026,
author = { Asimeng, Ernest AND Chen, Jun AND Mensah, Godbless AND Zhang, Jiaqi AND Han, Kai AND Chen, Yang AND Liu, Zhe},
title = { { PathoMamba: Piecewise-Diffeomorphic Registration via Stiffness-Modulated State Space Dynamics } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16889},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper introduces PathoMamba, a novel registration framework designed to address the correspondence-regularity conflict: healthy tissue requires high stiffness to prevent folding, whereas gliomas require high plasticity to model mass effects.
The primary contribution of this work is the Stiffness-Modulated Dynamics (SMD) mechanism. By reinterpreting the discretization step of a State Space Model (SSM) as learnable Numerical Inertia, the authors provide a way to spatially vary the network’s “memory” and feature update rate. This allows the model to treat healthy parenchyma as a rigid, high-inertia material and pathological tissue as a plastic, low-inertia material, and by employing the scaling-and-squaring algorithm, it guarantees topological safety.
The authors also propose the Therapy-Aware Biomechanical Loss to handle the heterogeneous nature of glioma evolution (simultaneous cavity collapse and recurrence).
- 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 reinterpretation of the Mamba discretization parameter as physical stiffness/numerical inertia is interesting and theoretically grounded. It moves beyond simple “mask-weighting” and embeds biomechanics into the architecture’s latent dynamics.
The use of an omnidirectional 6-way scanning mechanism successfully adapts the 1D nature of SSMs to 3D medical volumes. Integrating this with a Stationary Velocity Field (SVF) ensures diffeomorphic mappings, which is a clinical necessity.
The proposed method can achieve good topological folding performance while providing a 2x speedup over state-of-the-art Transformer-based models like TransMorph, highlighting the efficiency of the linear-complexity SSM backbone.
- 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.
Heavily relies on an external segmentation model (nnU-Net) to generate the Signed Distance Function (SDF) prior. While the Mamba backbone offers impressive inference speedups, the quantitative performance (1.88 mm mTRE) does not actually exceed existing state-of-the-art results from literature (e. g. , DIRAC-D or BiPyramid). The contribution appears to be purely about efficiency or the folding ratio, rather than a breakthrough in registration precision for the BraTS-Reg benchmark. The paper also lacks a rigorous reproduction of key state-of-the-art methods for the BraTS-reg benchmark, such as NR-IAM or DIRAC.
- 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.
- 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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
(1) A single 80/20 random split is insufficient for the heterogeneous nature of gliomas, necessitating k-fold cross-validation. Furthermore, the authors didn’t stratify landmarks into “near-tumor” versus “far-from-tumor” zones, which is critical for verifying the claimed biomechanical advantages of the model.
(2) The comparison primarily focuses on general baseline models rather than established BraTS-Reg state-of-the-art methods. The lack of reported accuracy improvement makes the contribution’s value unclear.
(3) The work is currently difficult to replicate due to a lack of source code.
- 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
This paper proposes a registration network for longitudinal glioma progression analysis. By incorporating a stiffness-modulated State Space Model (SSM) into a U-Net-like Mamba architecture, together with a mask-based Therapy-Aware Biomechanical Loss, the method aims to achieve relatively rigid registration outside the ROI (glioma region) while allowing deformable registration within and around the ROI. The method is well motivated for the specific clinical problem, as it encourages the registration to focus on pathological changes caused by glioma progression. The paper also evaluates the method on a public challenge dataset and reports promising registration accuracy together with diffeomorphic deformation.
- 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 proposes an interesting modification of Mamba by using a Signed Distance Function (SDF) to dynamically control the update behavior of the SSM in different regions. In this way, the healthy tissue can evolve more slowly while the tumour region can respond more quickly, which is well aligned with the characteristics of longitudinal glioma registration. 2.The paper further introduces a mask-based Trend-Aware Biomechanical Loss (TABL), which penalizes the deformation field according to tumour growth or shrinkage trends. This is a meaningful design for handling heterogeneous pathological changes and is more task-specific than a standard uniform regularization strategy. 3.The paper includes comparisons with several state-of-the-art methods on a public challenge dataset, and the ablation study also shows that the proposed components are useful. 4.The proposed method may also extend to other longitudinal studies with tumor-like lesions and mask guidance beyond glioma registration.
- 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 method section is not clearly described, and some important parts are missing or insufficiently explained, as detailed in the comments below. In addition, no source code link or code placeholder is provided. The authors should either describe the method in more detail or release the code to ensure reproducibility. 2.The experimental and results sections are also unclear. For example, important settings such as the LNCC window size and the number of scaling-and-squaring steps are not provided. Please see the comments below for more details. 3.All experiments lack statistical significance testing. For the BraTS-Reg state-of-the-art results collected from the literature, statistical testing may not be feasible, but at least the reproduced baselines should be tested. The same issue also applies to the ablation study.
- Please rate the clarity and organization of this paper
Poor
- 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.
- 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
1.The paper lacks the parameter settings for the scaling-and-squaring step, such as the number of integration/squaring steps. These details should be clearly described, or the authors should provide code for reproduction. 2.It would be better to add a short description of matrices A and B in Equation 1.3.In Equation 2, the variables W_img, W_sdf, and x_p. are not clearly described. Please add a brief explanation for each variable. 4.For L_sim, the authors use LNCC as the similarity loss, but LNCC implementations can vary, for example in box/window size. The authors should either provide a standard reference or release the code for reproduction. 5.The paper seems to mention L_MK, but its definition or introduction is missing. This should be clarified. 6.It is not clear how the OSM block is used within the U-Net-like architecture from both the figure and the text. The connection between the architecture figure and the method description should be made clearer. If space is limited, the authors could add a brief explanation and cite a reference. 7.In the ablation study, does the full method not include L_diff, or is it omitted in the table due to limited space? In addition, the configuration names are not well aligned with the method section. For example, “Input Concat” should be described more clearly so that readers can understand exactly which design it corresponds to. 8.In Fig. 2, the mask overlay seems to have no transparency, so the original warped image can barely be seen. The visualization should be improved to make the qualitative results easier to interpret. 9.Although the proposed pipeline relies on an auxiliary pretrained nnU-Net at inference to generate the pathology prior SDF, the authors have already briefly discussed this aspect. For a short conference paper, this is likely sufficient. It could still be useful for future work to further discuss robustness to segmentation errors and comparison with image-only baselines.
- 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 current manuscript still has substantial clarity issues, especially in the method description and implementation details. However, I find the core idea novel and well motivated, and it directly targets an important difficulty in longitudinal tumor registration. For this reason, I lean toward weak accept. If the authors can carefully address the above comments, particularly by making the method section clearer and improving reproducibility, the paper has the potential to become a good 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.
The authors provided appropriate explanations for several formulas and variables and promised to release the code after acceptance, supporting reproducibility. They also provided deeper explanations and reasonable justifications for the ablation study. Given that the proposed SDF-guided Mamba design and trend-aware biomechanical loss are well aligned with longitudinal glioma registration and may also be extended to other mask-guided tumour-like longitudinal studies, I recommend acceptance.
Review #3
- Please describe the contribution of the paper
The paper proposes PathoMamba, a deformable registration framework for longitudinal glioma imaging that explicitly addresses the conflict between enforcing smooth, topology-preserving deformations and capturing tumor-induced mass effects. Its key contribution is the reinterpretation of the State Space Model (SSM) discretization step as a learnable “numerical inertia”, which enables spatially varying stiffness by modulating the internal model dynamics rather than relying solely on loss-based constraints.
- 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.
A major strength is the novel reinterpretation of the SSM discretization step Δ as a learnable numerical inertia, enabling spatially varying stiffness directly in the model dynamics. This provides a principled mechanism to decouple rigid and non-rigid deformation, addressing the correspondence vs. regularity trade-off in longitudinal tumor registration . The proposed stiffness-modulated dynamics (SMD) go beyond prior mask-conditioned or loss-weighted approaches by actively modulating the evolution operator using anatomical priors (SDF). This shift from passive regularization to dynamics-level conditioning is technically meaningful. The therapy-aware biomechanical loss (TABL) introduces component-wise, physically interpretable constraints on volume change, improving modeling of heterogeneous tumor evolution beyond standard smoothness regularization. The architecture is efficient and well designed, mitigating SSM’s 1D bias via multi-directional scanning while maintaining linear complexity. Finally, the method achieves competitive accuracy with strict topological guarantees (0% folding) and is supported by targeted ablation studies, demonstrating effective integration of the proposed components
- 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 experimental validation is limited to a single benchmark (BraTS-Reg) with a fixed data split, and lacks evaluation across datasets or imaging conditions. This restricts conclusions about generalizability, especially given the strong dependence on tumor segmentation quality. 2.The method further relies on an auxiliary segmentation model (nnU-Net) at inference time, introducing an additional dependency that is not jointly optimized and may propagate errors. While some robustness analysis is reported, failure modes under inaccurate segmentations are not systematically studied. 3.the paper lacks comparisons to recent SSM-based registration models (e.g., MambaMorph, VMambaMorph) beyond basic baselines, making it difficult to assess whether the proposed stiffness modulation provides a consistent advantage over closely related architectures
- 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?
I recommend a weak accept because the paper presents a technically sound and well-motivated approach with a meaningful conceptual contribution, despite limited overall novelty. The key factor is the reinterpretation of the SSM discretization step as a learnable physical control variable, enabling spatially varying stiffness directly within the model dynamics. This provides a principled mechanism to address the correspondence–regularity trade-off and goes beyond standard loss-based constraint formulations . The method is coherent and well engineered, achieves competitive performance with strict topological guarantees, and is validated on a relevant benchmark. The weaknesse: primarily incremental novelty, reliance on segmentation priors, and limited experimental scope, prevent a stronger recommendation but do not outweigh the core contribution.
- 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 R1–R3 and AC. R2/R3 endorse our core contribution—reinterpreting the SSM step Δ as learnable Numerical Inertia—and all three acknowledge SMD, the SVF diffeomorphic guarantee, and TABL as novel.
[1] Notation (AC, R2). L_MK is the SECOND term of Eq.(4): (1/|Ω_shell|)·Σ ReLU(|J_φ|−1−δ), a Monro–Kellie peritumoral-shell penalty suppressing unphysical expansion (|J|>1+δ) in the healthy rim around the tumor; combined with L_diff (w≈1 there since D>0) this maintains |J|≈1, realizing intracranial volume conservation. Full objective: L_total = L_sim + λ_reg·L_diff + λ_bio·(L_TABL + L_MK). L_TABL applies a per-component Jacobian penalty driven by the observed log-volume ratio η_k = log(V_T1^k/V_T0^k): growing components (η_k>0) pushed toward |J|>1, shrinking ones toward |J|<1, robustly handling simultaneous recurrence and cavity collapse. W_img, W_sdf (Eq.2) are learnable 1×1×1 convs (per-voxel linear projections of x_p, D_p). A, B (Eq.1) are standard Mamba matrices [9]; A is S4D-real initialized, B input-dependent. LNCC uses a 9³ window (MONAI default); scaling-and-squaring uses 7 steps.
[2] Regularity and α (AC). We set α=1.0 with D_p in mm, so w(D)=σ(D): w(+5)=0.993, w(+1)=0.731, w(0)=0.500, w(−1)=0.269, w(−5)=0.007.The transition is smooth across a ~2mm band around the boundary—not a sharp drop. Three deliberate properties: (i) at the boundary w≈0.5 retains substantial smoothness, preventing folding (Tab.1: 0.00%); (ii) in deep healthy tissue (D≥5mm) w→1 enforces full rigidity where folding risk is highest; (iii) in the deep tumor core (D≤−5mm) w→0 admits the plasticity mass effect requires—this is intentional. Stability inside the tumor is additionally enforced by L_TABL and the SVF parameterization, so smoothness is never the sole constraint.
[3] Table 2 (AC, R2). Each row isolates one component. (1)→(2): adds passive SDF concatenation at the U-Net input (mask-conditioning baseline). (2)→(3): replaces passive concat with ACTIVE modulation of Δ via Eq.(2)—the proposed SMD; the sole architectural change, directly isolating dynamic vs. fixed Δ. (3)→(4): adds L_TABL+L_MK on top. L_diff is retained in EVERY configuration; the row-(4) “L_sim+L_TABL” label is shorthand for the newly added term—the full L_total objective applies to all rows. The (2)→(3) gap (~0.15–0.20mm) directly quantifies dynamic Δ over passive concatenation—our central architectural claim.
[4] SOTA positioning (R1, R3). The contribution is not raw mTRE but a unique combination: (i) competitive 1.88mm (tied with DIRAC-D; within 0.03mm of VMambaMorph 1.85); (ii) 0.00% folding—a strict topological safety guarantee no listed method matches; (iii) best tumor Dice 0.81; (iv) sub-second one-shot inference, 2× faster than TransMorph. LapIRN (1.82) and DIRAC-Opt (1.76) report MEDIAN TRE under instance optimization at 4–60s/case; mean vs. median and instance-opt vs. one-shot are not directly comparable. R3’s request for SSM comparison is already met in Tab.1: MambaMorph [10] (1.95), VMambaMorph [23] (1.85), and locally reproduced RegMamba (2.24) as the physics-blind ablation.
[5] nnU-Net robustness (R1, R3). §3.2 reports that under ±5mm SDF perturbation mTRE degrades marginally from 1.88 to ~1.94mm (the printed “0.6mm” is a transcription error), while folding remains 0.00%. Δ_p is co-modulated by image features (W_img·x_p in Eq.2), letting the network override mild segmentation errors via local texture. Catastrophic failure remains a known limitation, as stated.
[6] Other points. k-fold CV / multi-dataset / further reproductions (R1, R3) cannot be added at rebuttal per MICCAI rules; landmarks are strictly held-out and identical across all locally reproduced baselines, ensuring fair comparison on this split. Code will be released upon acceptance.
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.
This paper proposes an image registration approach for the BRATS-reg dataset. The goal of the proposed approach is to develop a registration approach that can be applied in the context of brain tumors and where one may want to account for different deformation patters within and outside the tumor region.
The proposed approach uses a state-space approach (Mamba) to modulate the behavior of the network when predicting a stationary velocity field (SVF). Specifically, it makes use of a tumor segmentation and its associatd signed distance field to allow for rapid state changes inside the tumor region and for comparatively smaller changes in the surrouding healthy tissues. Beyond this there is a weighted diffusion regularization of the SVF, a trend-aware biomechanical loss, an LNCC similarity measure, and an undefined L_{MK} loss term.
The paper has several strength:
- The reviewers appreciated the goal of using an active modulation of the Mamba timestep to model faster and slower changing tissue effects as well as the integration with the tumor segmentation.
- The reviewers appreciated the Trend-Aware Biomechanical Loss.
- Reviewers appreciated that due to the SVF formulation diffeomorphic transformations could be obtained.
The paper has several shortcomings:
- The work is very task-specific and hence is only evaluated on one dataset (BRATS-Reg)
- Some of the notation is not well decribed. (L_{MK}, W_{img}, W_{sdf})
- There is significant lack of clarity in the model descriptions themselves as well as in the experiments (see the detailed comments by reviewer 2).
- No real accuracy improvements over state of the art are achieved.
- The method relies on an external segmentation network.
For the rebuttal please focus on the following aspects:
- Please clarify how you will fix the notation / descriptions. For example, what is L_{MK}? What are W_{img}, W_{sdf}? (Just learned parameter matrices?)
- Provide a discussion on what L_{TABL} and L_{MK} do.
- Address reviewer questions regarding comparisons to state of the art approaches on this dataset.
- Provide a clearer discussion of what Table 2 shows and how one should interpret these results. For example, are the differences between methods (2) and (3) in this table simply your approach with the dynamic timestep \Delta versus keeping it fixed (as this seems to be the key contribution)? And is the difference between 3 and 4 the addition of L_{TABL}? What happened to L_{diff}?
- Please also comment on the achieved regularity of the velocity field / deformation map inside and outside the tumor region. In particular, what is alpha in Eq. (3) set to and how much lower does the weight become in the tumor region? If it drops very fast it seems there will be hardly any regularity constraint (seems potentially bad). If it drops too slow then this would encode less regularity the deeper one is inside a tumor (is this the intended behavior?). Please discuss.
- Please address any other reviewer concerns as space permits.
- 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 an image registration approach for the BRATS-reg dataset. The goal of the proposed approach is to develop a registration approach that can be applied in the context of brain tumors and where one may want to account for different deformation patters within and outside the tumor region.
The proposed approach uses a state-space approach (Mamba) to modulate the behavior of the network when predicting a stationary velocity field (SVF). Specifically, it makes use of a tumor segmentation and its associatd signed distance field to allow for rapid state changes inside the tumor region and for comparatively smaller changes in the surrouding healthy tissues. Beyond this there is a weighted diffusion regularization of the SVF, a trend-aware biomechanical loss, an LNCC similarity measure, and an undefined L_{MK} loss term.
The paper has several strength:
- The reviewers appreciated the goal of using an active modulation of the Mamba timestep to model faster and slower changing tissue effects as well as the integration with the tumor segmentation.
- The reviewers appreciated the Trend-Aware Biomechanical Loss.
- Reviewers appreciated that due to the SVF formulation diffeomorphic transformations could be obtained.
The paper has several shortcomings / questions:
- The work is very task-specific and hence is only evaluated on one dataset (BRATS-Reg)
- Some of the notation is not well decribed. (L_{MK}, W_{img}, W_{sdf}) though could be in a revised version (and was addressed in the rebuttal).
- There is significant lack of clarity in the model descriptions themselves as well as in the experiments.
- No real accuracy improvements over state of the art are achieved. Though the rebuttal argues that the approach has other benefits such as 1) competitive accuracy, 2) no folding, 3) excellent tumor Dice, and 4) sub-second one-shot inference.
The rebuttal clarified some of the questions / shortcomings above and the reviewers are largely positive. This meta reviewer is concerned about clarity of the writing.
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.
After reading the manuscript and following the review and rebuttal process, I believe this is a strong contribution for the MICCAI audience. The problem addressed is highly challenging, and the proposed solution is innovative. I particularly appreciated the Stiffness-Modulated Dynamics (SMD) mechanism and its integration into a state-space framework. I generally agree with the reviewer’s comments, and I believe the manuscript would further benefit from addressing the most relevant questions and concerns they raised. Incorporating these clarifications would strengthen the overall quality of the paper.
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.
The paper received mixed recommendations following the rebuttal. Reviewer 2 upgraded their assessment to Accept after the authors clarified the formulation, committed to releasing the code, and provided additional hyperparameter details. Reviewer 3 maintained a Weak Accept recommendation, while Reviewer 1 did not update their review and remained at Weak Reject.
Although Reviewers 2 and 3 offered generally positive evaluations and acknowledged the novelty of the proposed method, both continued to express substantive concerns regarding the clarity of the presentation and the limited scope of the experimental validation. In addition, the method’s reliance on segmentation priors and the relatively modest gains over prior state-of-the-art methods further weakened the case for acceptance. In light of the post-rebuttal discussion and the overall reviewer feedback, I recommend rejection.