Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Recent developments in neural networks have improved deformable image registration (DIR) by amortizing iterative optimization, enabling fast and accurate DIR results.
However, learning-based methods often face challenges with limited training data, large deformations, and tend to underperform compared to iterative approaches when label supervision is unavailable. While iterative methods can achieve higher accuracy in such scenarios, they are considerably slower than learning-based methods. To address these limitations, we propose VoxelOpt, a discrete optimization-based DIR framework that combines the strengths of learning-based and iterative methods to achieve a better balance between registration accuracy and runtime. VoxelOpt uses displacement entropy from local cost volumes to measure displacement signal strength at each voxel, which differs from earlier approaches in three key aspects.
First, it introduces voxel-wise adaptive message passing, where voxels with lower entropy receives less influence from their neighbors.
Second, it employs a multi-level image pyramid with 27-neighbor cost volumes at each level, avoiding exponential complexity growth.
Third, it replaces hand-crafted features or contrastive learning with a pretrained foundational segmentation model for feature extraction.
In abdominal CT registration, these changes allow VoxelOpt to outperform leading iterative in both efficiency and accuracy, while matching state-of-the-art learning-based methods trained with label supervision. The source code will be available at https://github.com/tinymilky/VoxelOpt.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3887_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/tinymilky/VoxelOpt

Link to the Dataset(s)

https://learn2reg.grand-challenge.org/Datasets/

BibTex

@InProceedings{ZhaHan_VoxelOpt_MICCAI2025,
        author = { Zhang, Hang and Zhang, Yuxi and Wang, Jiazheng and Chen, Xiang and Hu, Renjiu and Tian, Xin and Li, Gaolei and Liu, Min},
        title = { { VoxelOpt: Voxel-Adaptive Message Passing for Discrete Optimization in Deformable Abdominal CT Registration } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15963},
        month = {September},
        page = {667 -- 677}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes VoxelOpt, a discrete optimization framework for image registration that leverages feature maps from a foundational segmentation model and employs entropy-based adaptive Gaussian smoothing to solve decoupled convex subproblems. VoxelOpt achieves high accuracy in inter-patient abdominal CT registration using a multi-level image pyramid via iterative optimization, without requiring supervised training/learning.

  • 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. VoxelOpt makes use of feature maps from pre-trained foundational segmentation model, which results in smoother feature entropies (more low-entropy voxels) within organs compared to MIND and raw images
    2. This work resembles adaptive message passing in mean-field inference when solving the regularization subproblem. It proposes to weight the Gaussian smoothing with voxel-wise entropy (computed from the probability map derived from the cost volume),with the intuition that strong displacement signals (low entropy) should send strong messages to neighbouring uncertain voxels (high entropy) and receive few updates (similar to control points)
    3. VoxelPrompt enable large displacements capture using 26 neighbors at each level with a 5-level Laplacian image pyramid, and it use scaling and squaring to approximate diffeomorphic registration
    4. The structure of the result tables is informative, and the ablation studies are well designed and clearly presented
  • 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. Incremental contribution over ConvexAdam[1]. The main additions over ConvexAdam are the use of foundational model features and the entropy-based adaptive Gaussian smoothing
    2. Potential Dice score gain from the foundation segmentation model (possible information leakage). Comparing #2 and #4 in Table 2 shows a significant improvement when switching MIND to foundation feature maps. Since the foundation model is explicitly trained for organs segmentation, it might provide organs information during registration, explaining the Dice improvement
    3. Inter-patient Abdominal CT registration might be less clinically relevant. The numbers are reported without standard deviation nor significance test. Besides, the average Dice (roughly 60) might be sub-optimal compared to segmentation performance. A per-organ Dice analysis would help clarify performance across anatomically distinct structures.
    4. Lack of qualitative results. Figure 1 effectively illustrates the core contribution (entropy visualization), but no registered images, segmentations, or deformation fields are shown. These might also offer important qualitative insights and support evaluation.
    5. Minor: The variable qb in the last 3 lines of page 4 is not defined

    [1] Siebert H, Großbröhmer C, Hansen L, Heinrich MP. ConvexAdam: Self-Configuring Dual-Optimisation-Based 3D Multitask Medical Image Registration. IEEE Transactions on Medical Imaging. 2024 Sep 16.

  • 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.

  • 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
    1. A more detailed explanation or intuition behind the entropy-to-smoothing mapping function σ(x)\sigma(x)σ(x) (Eq. 6) would strengthen the contribution of the entropy-based adaptive Gaussian smoothing.
    2. Additional experiments analyzing the impact of the foundational segmentation model’s features would be valuable. Using prior knowledge from a pre-trained model is a strong idea (who can say no to a free lunch), but it is important to clarify its precise contribution to performance gains.
  • 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 — could be rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Overall, this is a very interesting and well-motivated paper with solid methodology. However, the novelty and significance of the core contributions—relative to existing work such as ConvexAdam—should be more clearly justified and quantified to establish a stronger case for impact.

  • 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 describes a discrete optimisation based framework for registering medical images by using a multi-level (feature) pyramid with a restriction to 26 neighbours for cost volumes / message passing optimisations each.

  • 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 introduces several new concepts into discrete optimisation for 3D registration: 1) voxelwise entropy, 2) voxelwise adaptive message passing and 3) use of foundational feature extractors. It achieves strong performance compared to SOTA on the challenging task of inter-subject abdominal CT registration. The experimental validation is extensive and the paper is generally well written. The ablation study reveals an interesting aspect, namely that kernel size 1 (26 neighbours) for the cost volume is superior to larger ones. Even without foundational features, the method delivers solid performance (50% Dice).

  • 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.

    I only have rather minor weaknesses: the number of iterations for the u-v relaxation are fixed to 6 and not evaluated. The difference in loss function for semi-supervised models (I assume Dice overlap on the 13 abdominal labels) is not clearly mentioned leaving some confusion - also LapIRN or ConvexAdam would yield slightly higher scores than RDP in this case. In future work a second dataset should be considered. The foundational feature training seems to include the same dataset as used for evaluation, which is not ideal. The smoothness of the transformation (which is expected to increase to some degree with more complex alignment) is not as good as e.g. Deeds.

  • 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?

    The method tackles a challenging problem and achieves SOTA results with a novel formulation of discrete optimisation based registration.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #3

  • Please describe the contribution of the paper

    VoxelOpt directly attacks the problem of image registration by aligning features from a pretrained segmentation model, and a clever message passing algorithm to route good alignments from featureful voxels into difficult-to align featureless regions.

  • 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 non-iterative registration via pyramid + global convex solver is very interesting, especially without need for training. This allows very rapid registration. The entropy-based message passing formulation is well justified mathematically, which likely saved an enormous amount of hyperparameter tuning work.

  • 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 method appears to be general, but is only evaluated on abdominal CT registration. The method is presented as an unsupervised method, but supervision was used to train the feature extractor which is based on a segmentation model. A comparison to ConvexAdam+nnUNet is missing, which is unfortunate as that technique is very similar to the presented approach.

  • 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?

    The mathematical formulation presented appears to be correct, and performs well at abdominal registration. The baselines chosen are strong and the method outperforms them significantly. The writing is exceptionally clear.

  • 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

N/A



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



back to top