Abstract

Registration of diffusion MRI tractography is an essential step for analyzing group similarities and variations in the brain’s white matter (WM). Streamline-based registration approaches can leverage the 3D geometric information of fiber pathways to enable spatial alignment after registration. Existing methods usually rely on the optimization of the spatial distances to identify the optimal transformation. However, such methods overlook point connectivity patterns within the streamline itself, limiting their ability to identify anatomical correspondences across tractography datasets. In this work, we propose a novel unsupervised approach using deep learning to perform streamline-based dMRI tractography registration. The overall idea is to identify corresponding keypoint pairs across subjects for spatial alignment of tractography datasets. We model tractography as point clouds to leverage the graph connectivity along streamlines. We propose a novel keypoint detection method for streamlines, framed as a probabilistic classification task to identify anatomically consistent correspondences across unstructured streamline sets. In the experiments, we compare several existing methods and show highly effective and efficient registration performance. interpretable keypoints that effectively represent whole-brain streamlines without relying on any ground truth annotations. For both streamline-based and volumetric metrics, our approach has achieved a better performance than other popular learning-based and optimization-based methods with fewer parameters and reduced computational time.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3289_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)

N/A

BibTex

@InProceedings{WanJun_ANovel_MICCAI2025,
        author = { Wang, Junyi and Du, Mubai and Wu, Ye and Li, Yijie and Wells III, William M. and O’Donnell, Lauren J. and Zhang, Fan},
        title = { { A Novel Streamline-based diffusion MRI Tractography Registration Method with Probabilistic Keypoint Detection } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15971},
        month = {September},
        page = {23 -- 33}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    Authors propose a novel unsupervised approach using deep learning to perform streamline-based dMRI tractography registration.

  • 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.

    This paper introduces an unsupervised approach using deep learning for dMRI tractography registration. Overall, this is a novel method to register streamlines by finding the anatomical keypoints and applying the optimized transformation.

  • 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.

    Authors should mention the major findings towards the end of the abstract instead of the general sentence: “In the experiments, we compare several existing methods and show highly effective and efficient tractography registration performance.” Please quantify effective and efficient.

    -Registration computation time is not mentioned in the paper. This will help in commenting about the time taken to register.

    -It will be helpful to write some details about the ABD and wDice metrics. For example, lower the better ABD. Also, authors should discuss how significant is 0.5 improvement in ABD. Given, there is no improvement in wDice. To me, SynthMorph method performs comparable to the proposed method for ABD and outperforms the proposed method for wDice by 5%. Authors should explain the merits of their results and highlight the advantage of the proposed method in comparison with the methods in Table 1.

    -Figure captions are poorly written. Please elaborate on the observations in the Figures. For example, Fig2., please write in detail what we are seeing in the figure.

    -I am curious how the results would be if compared to a basic unsupervised pipeline which is very commonly used: register one image to another -> save the transformation matrix -> apply the same transformation to the streamline points. If the authors can visualize and interpret the differences between these methods, that would make the paper stronger.

  • 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 provide sufficient information for 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.

    (4) Weak Accept — could be accepted, dependent on rebuttal

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

    The lack of quantification of the effectiveness of the proposed method and not providing the code to reproduce the results make me incline towards weak accept.

  • 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 #2

  • Please describe the contribution of the paper

    This work proposes a two-step framework that first identifies tract keypoints and subsequently aligns tracts using correspondence information. However, concerns arise due to insufficient clarity regarding certain technical 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.
    1. This work introduces a keypoint-based registration approach tailored specifically for tractography registration, operating directly on streamline data.
    2. The experimental results demonstrate the effectiveness of the proposed keypoint detection strategy, validating the proposed 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. Dataset Selection and Real-Data Generalizability The dataset used in this study appears carefully selected streamlines, consisting of 105 subjects, yet the experiments only involve 100 subjects. It is unclear why five subjects were excluded. Additionally, the paper does not clearly address how the proposed method performs on other real-world datasets, raising concerns about its broader generalizability.

    2. Ambiguity in Graph Construction The manuscript lacks clarity regarding how the graph structures are built from individual streamlines or neighboring streamline points. Is this line-type graph, 10.1007/978-3-030-35817-4_11? The statement, “The network identifies keypoint locations by modeling their streamline point connectivity relative to the neighboring streamlines rather than relying solely on their spatial coordinates,” is ambiguous and requires detailed explanation.

    3. Unclear Definition and Advantage of Edge Convolution The definition and implementation details of edge convolution on graphs are insufficiently explained. Specifically, the manuscript does not clearly state how edge convolution differs from node convolution and what specific advantages it provides in the context of tractography registration.

    4. Sampling Strategy Ambiguity The manuscript does not adequately describe how streamline patches are randomly sampled from fixed and moving tractography datasets. Furthermore, it remains unclear whether the method could perform effectively under a purely random sampling strategy, and whether there is any specific criterion for patch selection.

    5. Insufficient Figure Captions and Descriptions Figure captions and accompanying text descriptions are insufficient to fully interpret the presented figures. The meaning of colors, points, and lines shown in figures should be explicitly clarified to enhance readability and understanding.

    6. Clarification on Metric Calculation (ABD and Dice) The Dice scores reported for SynthMorph appear superior compared to other methods, raising the question of how exactly the Average Bundle Distance (ABD) and Dice metrics were computed. The manuscript should clearly specify whether these metrics rely on streamline coordinates or voxel-level representations.

    7. Typographical and Formatting Errors The manuscript contains several typos and minor errors, particularly within equations and main text descriptions, which reduce overall readability and professionalism. Careful proofreading is recommended to resolve these issues.

  • 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 provide sufficient information for 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.

    (4) Weak Accept — could be accepted, dependent on rebuttal

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

    The authors propose an interesting framework for streamline registration. However, several issues let me cannot make the decision from the current manuscript.

  • 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

    Authors proposed an effective method for streamline registration without volumetric reference. It first extracts keypoints from moving and fixed tractograms, and estimates a deformation field to align two groups of streamlines. Their method showed better performance than existing volumetric and streamline-based registration methods. It adds a good tool for efficient group-level white matter fiber analysis.

  • 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 framework is well-designed and efficient.
    2. Extracting streamline landmarks to assist registration is a good way to reduce computational cost since there could be tremendous redundant streamlines generated in whole brain tractography.
    3. Experiments are solid. Major methods including volumetric and streamline-based methods are compared, compared to which the method showed good 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.

    If possible, I would suggest that the author conducts student’s t-test to add significance of their experimental results. The test could be performed registration results across the proposed method and existing methods.

  • 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?
    1. The framework is well-motivated and efficient.
    2. The proposed method is a new and effective tool for group-level white matter fiber registration and outperformed existing methods.
    3. The targeted problem is important and challenging in the dMRI tractography field.
    4. The paper is well organized and easy to follow.
  • 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 sincerely thank all the reviewers for their valuable feedback on our paper. We will thoroughly revise the manuscript to address the reviewers’ comments prior to publication. We will make the code publicly available upon acceptance. For specific comments:

R1: The reviewer noted that registration computation time was not reported. We appreciate this observation. We have computed the registration time: on average, our method takes about 15s to register all 1.6M moving streamlines to the fixed space. We will include this timing information in the revised version of the manuscript.

Regarding the improvement in ABD, we acknowledge that the numerical difference may appear modest. However, we note that the ABD values across most of the compared methods vary within a narrow range—typically around 0.5mm. In this context, an improvement of about 0.13mm is a relatively meaningful gain in registration accuracy at the tract level. As for the higher wDice performance of SynthMorph, we believe this is primarily due to its volume-based training, which inherently favors voxel-level metrics like wDice. However, wDice does not capture streamline connectivity and therefore cannot reflect the geometric and topological properties of the tracts.

Regarding the suggestion to compare against a basic unsupervised pipeline: indeed, volume-based methods such as SyN and SynthMorph follow such pipelines. We will make this point clearer in the revised experimental section.

R2: We appreciate the reviewer’s suggestion to include additional paired t-tests. If permitted, we will include the t-test results in the revised version.

R3:The reviewer noted that we used 100 out of 105 subjects from TractSeg and did not include other datasets. We apologize for the oversight—this was a typo. In fact, we used all 105 subjects from the TractSeg dataset. We will correct this in the manuscript. Regarding the use of other datasets: We agree with the reviewer that it is important to evaluate our method on datasets generated with different tractography algorithms, and we will include a discussion about this in future work.

The reviewer also raised questions about our graph construction and the use of edge convolution. To clarify: our graph is constructed solely based on individual streamlines. Each node represents a point on a streamline and stores its 3D coordinates. Edges exist only between consecutive nodes on the same streamline—there are no edges connecting different streamlines. The use of edge convolution allows the neural network to learn not only from local point features but also from their structural connectivity, making it topology-aware. We will add this explanation to the manuscript.

Lastly, regarding the random streamline selection strategy, we emphasize that no selection tricks were introduced. In each training batch, we randomly sampled 2.2K streamlines from the full set of approximately 1.6M. These 2.2K streamlines generally cover all parcels in TractSeg and provide a good approximation of whole-brain streamlines.




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