List of Papers Browse by Subject Areas Author List
Abstract
Accurate and automatic lifespan brain cortical surface reconstruction (CSR) is crucial for analyzing brain development and aging. Traditional pipelines involve multiple processing steps, which are time-intensive and inefficient for handling larger datasets. While deep learning-based methods can accelerate reconstruction speed and produce high-quality meshes compared to traditional approaches, they are often constrained to a single time point. The limitation arises from the significant variations in cortical surfaces across age groups, particularly in folding patterns. In this paper, we propose a novel curvature-guided diffeomorphic mesh deformation framework for lifespan brain CSR. Specifically, to preserve correct topology structure and uniformity, the framework employs multiple deformation blocks to gradually warp a simple smooth template mesh to a complex target surface with high folding. Considering that curvature is closely associated with folding patterns, we introduce curvature map prediction as an auxiliary task to guide the deformation process, enhancing the anatomical accuracy to facilitate subsequent cortical morphometry. Notably, incorporating curvature can also expedite model convergence. Our method is evaluated on a large-scale brain dataset with 2,132 subjects spanning ages 0 to 100 years. Experimental results show that our reconstructed surfaces have fewer geometric errors and optimal mesh regularity while being several orders of magnitude faster than traditional pipelines. Our code is available at https://github.com/TL9792/CCF.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1603_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{TenLin_ACurvatureGuided_MICCAI2025,
author = { Teng, Lin and Zhao, Shen and Shi, Feng and Shen, Dinggang},
title = { { A Curvature-Guided Diffeomorphic Mesh Deformation Framework for Lifespan Brain Cortical Surface Reconstruction } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15960},
month = {September},
page = {13 -- 23}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper presents an enhanced cortical surface reconstruction pipeline building upon CorticalFlow++. The improvements focus on both convergence efficiency and reconstruction quality. The authors introduce two key contributions: incorporating tissue segmentation masks as input, and integrating curvature information as a prior to guide the reconstruction process.
- 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 technical submission tackles a fundamental challenge in brain morphometric analysis (cortical surface reconstruction). A key strength of the paper lies in its practical improvements to the CorticalFlow++ pipeline, enhancing both performance and applicability. The novelty stems from two main aspects: the integration of tissue segmentation masks derived from a segmentation model, and the use of curvature information as a prior. These enhancements contribute to more accurate and stable reconstructions, particularly in highly convoluted regions of the cortex, such as sulci and gyri, where many state-of-the-art methods tend to underperform. Additionally, the proposed method demonstrates competitive or superior performance against several leading approaches, highlighting its robustness and clinical relevance.
- 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 improvements introduced in this work, while effective, appear to be incremental rather than fundamentally novel. 2) The use of a tissue segmentation mask as input introduces additional information beyond the raw MRI, which raises important considerations. Firstly, the paper references a “proposed segmentation model” in the introduction, but this model is not clearly defined or described. For the sake of reproducibility and transparency, it is crucial to specify which segmentation model is used and whether it is publicly available or trained in-house. Secondly, the quality and reliability of the reconstruction will inherently depend on the quality of the segmentation. This dependency raises concerns about the generalizability of the method—particularly on lower-resolution MRI scans or in clinical settings where segmentation performance may degrade. It would be important to clarify whether only high-quality scans were used, as this could limit the scalability and robustness of the approach in real-world applications.
3) The ablation study, which is essential to understanding the individual contributions of the proposed components, is unfortunately relegated to the supplementary material. Based on the qualitative description provided, it appears that the majority of the performance gain stems from the tissue segmentation mask rather than the curvature prior. This is somewhat surprising, as the curvature-based prior seems to be the more novel and technically interesting contribution. A more detailed and accessible analysis of the ablation results in the main paper would help clarify the respective impact of each component.
4) Additional discussion around the choice of key hyperparameters would strengthen the paper. For instance, the rationale behind using only three iterations of the U-Net is not explained—was this empirically optimal, or driven by computational constraints? Similarly, the selection criteria for the number of mesh vertices is not discussed. Providing insight into these choices would help readers better understand the design decisions and potentially reproduce or adapt the approach for other settings. Also, how is chosen the initial template mesh to fit all the others?
- 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.
- 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
There is a small typo in the last sentence: “n contrast, our method…” should be corrected to “In contrast, our method…”.
Additionally, the quantitative results indicate that CorticalFlow++ [17] underperforms compared to what was originally reported in its paper. It would be valuable to evaluate your pipeline on the same dataset used in [17] (OASIS3) to provide a direct comparison and clarify whether the performance differences are due to dataset variations, implementation differences, or other factors.
- 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?
As written above, the improvements introduced in this work, while effective, appear to be incremental rather than fundamentally novel.
- 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.
After reading the rebuttal and considering the other reviews, I decided to increase my score. The rebuttal clarified the role and implementation details of the segmentation model, addressed concerns about generalization to lower-resolution data, and acknowledged ongoing efforts to extend applicability to raw MRI scans. The clarification and quantitative results from the ablation study highlight that the curvature prior is indeed a core technical innovation, with substantial impact on reconstruction quality, especially in highly folded cortical regions. The authors’ plan to move key ablation results into the main paper addresses concerns about visibility and accessibility of these findings. Overall, the rebuttal resolves the main concerns.
Review #2
- Please describe the contribution of the paper
Generating a representation of the cortical surface is an important intermediate processing step in many image-analytic workflows. This submission proposes a mesh deformation framework for generating said surfaces that uses curvature to guide the deformation and promises to maintain mesh topology. The method is compared against similar approaches and evaluated on a large-scale database compiled from several publicly available sources.
- 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 topic of this submission is scientifically interesting and of potential interest to the audience of this conference. The text is mostly well written (except where noted below) and straightforward to understand for a reader with a background in recent approaches for neuroimaging data analysis.
- 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.
A few, rather minor issues should be considered:
- Several conventional frameworks for diffeomorphic registration/deformation were developed over the last 20 years. Please, convince the skeptical reader of any advantages of reproducing this work in the CNN domain.
- It is understood that this framework does not guarantee a diffeomorphic transformation and “tries to avoid” self-intersections. Please, include some quantitative information how often self-intersections are detected in the output.
- Any reference to the supplementary information should be removed, as authors decided to retract that information.
- Add bibliographic detail to Refs. 4, 8, 15, and 20.
- 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.
- 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?
This is considered as a well-researched area. The advantage of re-inventing nonlinear registration in a CNN framework is unclear.
- 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 #3
- Please describe the contribution of the paper
This paper proposes a curvature-aware deep learning framework for diffeomorphic mesh deformation to reconstruct the cortical surface from T1-weighted MRI. The method deforms a template mesh using an ODE-based approach, which preserves the topological properties of the resulting surface. Experimental results demonstrate that the proposed method outperforms several existing approaches, particularly for brain development 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.
This paper builds upon the existing CorticalFlow++ framework, ensuring availability and reproducibility. The theoretical foundation and network architecture are straightforward and easy to follow, which contributes to the clarity and accessibility 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.
Since this work builds upon the existing CorticalFlow++ framework, the methodological novelty is somewhat limited. A more comprehensive comparison with other recent methods, such as Vox2Cortex, would strengthen the evaluation and better support the claimed performance improvements.
- 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.
- 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
I recommend that the authors clarify the explanation in Section 2.1 and Figure 1. According to the text, Ti is the input to the DMD module and Si is the output. However, the flowchart in Figure 1 indicates that Ti is both the output of the DMD and the input to the next DMD, which is inconsistent and confusing.
- 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?
This paper aims to improve the accuracy of cortical surface reconstruction, which is an important task in brain anatomy analysis. The architecture of the proposed method is interesting and clearly explained. Experimental results demonstrate performance improvements. However, as the method is built upon an existing framework, the overall novelty is limited.
- 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.
In the rebuttal, the authors highlight the application of their method to lifespan-wide anatomical reconstruction and the use of curvature features to enhance performance. They also acknowledge the need for broader comparisons with existing methods and indicate plans to address this in future work. Given these clarifications and the potential impact of the proposed approach, I recommend accepting the paper.
Author Feedback
We thank the AC and all reviewers for their constructive feedback. Please find below our point-by-point responses.
- Advantage and Novelty (R1 & R2 & R3) Our work addresses a long-standing challenge in cortical surface reconstruction – accurately modeling highly folded cortical regions across a wide age range. Traditional methods like FreeSurfer rely on multi-iterative optimization starting from an initial mesh, which are computationally expensive and often adult-specific. In contrast, CNN-based methods: (1) drastically reduce runtime from hours to seconds per subject; (2) adapt to lifespan-wide anatomical variabilities; and (3) is scalable for incorporating priors to improve the accuracy. Building upon an existing framework, we introduce crucial enhancement (e.g., curvature) that significantly improve anatomical fidelity, as validated quantitatively and qualitatively. This represents a meaningful and practical advancement, particularly for fine-grained, lifespan-wide brain analysis.
- Evaluation of Self-intersections, Robustness and Generalization (R1 & R3) To assess topological validity, we report the percentage of self-intersecting faces (%SIF), achieving 0.31%/0.73% for inner/outer surface reconstruction, respectively. Compared to our baseline (CorticalFlow++), our method presents respective improvements of 28% and 81%. Our method is robust to segmentation maps from various sources. For low-resolution data, we apply our prior super-resolution technique before segmentation and reconstruction. While better segmentations improve accuracy, our method remains effective after preprocessing. This demonstrates strong potential for clinical applicability. Further evaluate on raw MRI scans is planned, and we acknowledge that surface reconstruction from low-resolution data is an important direction for future work. Our experiments also reveal that CorticalFlow++ performs well on adult data but degrades on infant data. We also compared performance on the OASIS3 dataset. Our method outperforms CorticalFlow++ with 10%/18% (CD) and 8%/9% (HD) improvements for inner/outer surface reconstruction, confirming that the performance gap stems from methodological differences rather than dataset bias.
- Alation Analysis (R3) Our ablation shows that the curvature prior yields 67%/69% (CD) and 36%/33% (HD) improvements for inner/outer surface over CorticalFlow++, respectively. Compared to CorticalFlow+++Itissue, improvements are 18%/21% (CD) and 12%/13% (HD). This demonstrates that while segmentation masks aid reconstruction via clearer boundary, curvature priors are the key for accurate modeling of high-folding regions and contributes significantly to structural fidelity. Key ablation results in Supplementary Material will be moved in the final paper.
- Technical Clarifications (R2 & R3) We apologize for the confusion in Section 2.1 and Figure 1. Ti is the input to the DMD module; Ti+1 is passed to the next DMD module. The Figure 1 will be updated accordingly to clearly depict this flow. Our segmentation model is an in-house trained network composed of multiple CNN blocks, as supported in our prior work.
- Hyperparameter Justification (R3) Three iterations were selected empirically, as more iterations increased cost without improving accuracy. The number of vertices follows CorticalFlow++, which was empirically validated to balance anatomical precision and computational efficiency. The initial template mesh is a tight genus-zero surface generated from the union of signed distance fields of training meshes. This designed choice will be further studied in future work.
- Minor Concerns (R1 & R2 & R3) We will remove all references of the supplementary material and include full citation details for Refs. 4, 8, 15, and 20. We agree with R2 that future work should include comparisons with Vox2Cortex and CortexODE. The typo “n contrast” will be corrected to “In contrast”.
All updates and references will be included in the final version.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
The authors did a good job of rebuttal. This paper received three positive ratings from the reviewers. The AC concurs with the reviewers’ evaluations, and this paper is ready to be accepted at MICCAI’25.