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

Abstract

We present a spherical CNN-based interactive segmentation framework that leverages a shape-adaptive guidance signal for labeling cortical sulci in the lateral prefrontal cortex (LPFC). Recent neuroscience studies have examined the relationship between sulcal morphology and human cognition, focusing on small and shallow sulci that emerge late in gestation. Due to their variable locations and shapes, these sulci are identified manually by trained raters, which is a labor-intensive process that limits sample sizes in analyses. Although spherical CNN-based automatic sulcal labeling has shown reasonable accuracy for large, consistent sulci, labeling small and variable sulci remains challenging and often requires manual validation and correction. While deep interactive segmentation bridges the gap between manual and automatic labeling, user interactions should be carefully encoded into guidance signals in order to guide the models to accurate refinement. Here, we propose a shape-adaptive guidance signal on the spherical surface, in which the signal is encoded by solving the eikonal equation with a speed function based on mean curvature of the original surface. We compare our method with other guidance signals by training models to label 17 deep and shallow LPFC sulci. Our shape-adaptive signal demonstrates strong initial-click accuracy in small sulci and maintains effective iterative refinement with additional clicks. The code is publicly available at https://github.com/Shape-Lab/ISUS.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3242_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Shape-Lab/ISUS

Link to the Dataset(s)

N/A

BibTex

@InProceedings{SonJiw_ShapeAdaptive_MICCAI2026,
        author = { Son, Jiwon AND Willbrand, Ethan H. AND Parker, Benjamin J. AND Weiner, Kevin S. AND Lyu, Ilwoo},
        title = { { Shape-Adaptive Guidance Signal for Interactive Cortical Sulcal Labeling } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16894},
        month = {September},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    Providing efficient tools for supporting expert-based annotations, especially, in the context of developing ground truth data bases. This submission describes an approach for supporting an expert in labeling secondary and ternary sulcal structures in the lateral prefrontal cortex. A local encoding of local structural properties is used to guide an expert-chosen location to its optimal position.

  • 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 within the scope of this conference and of potential interest to its audience. The work described here is an incremental but considerable advance over previous approaches. The text is straightforward to understand for a reader with knowledge in recent approaches for image 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.

    Why did you choose mean curvature over geodesic depth? Using depth might be smoother in space.

  • 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?

    This is an incremental but potentially useful advance.

  • 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 paper presents an interactive framework for cortical sulcal labeling on spherical representations of the cortical surface. While framed as a labeling task, the method effectively performs binary segmentation per sulcus, using a spherical CNN (SPHARM-Net) combined with a novel shape-adaptive guidance signal. User interactions (clicks) are encoded via a weighted geodesic distance transform (WGDT), computed by solving an eikonal equation with curvature-dependent propagation, allowing the signal to follow cortical folding patterns. The method is evaluated on 17 sulci in the lateral prefrontal cortex (LPFC) using HCP data. Results show improved Dice scores over automatic baselines and conventional guidance signals (ADT, Disk), particularly for small and highly variable sulci, and especially in low-interaction regimes (e.g., single click).

  • 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.

    Novel and elegant methodological contribution: The use of curvature-driven propagation via the eikonal equation to encode user interactions is original and well-motivated. Clear improvement over baseline guidance signals: The geometry-aware WGDT signal shows advantages over equidistance-based approaches. Technically sound framework: The integration of spherical CNNs, interactive refinement, and PDE-based propagation is coherent.

  • 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.

    Evaluation setting: The interactive process relies on simulated clicks derived from ground truth (largest error region), representing an oracle scenario. This assumes optimal user behavior and likely overestimates real-world performance. It would be important to evaluate and discuss the effect of incorrect or suboptimal user clicks. Robustness: No analysis of incorrect or noisy user clicks, which is critical for interactive systems. Runtime: No reporting of computational cost or interaction time (including solving the eikonal equation), making it difficult to assess practical usability compared to automatic methods. Generalization: The evaluation is limited to a relatively small and homogeneous dataset, consisting mainly of healthy subjects within a narrow age range. Therefore, the applicability to and validation on different age groups or pathological populations, where sulcal morphology may vary, remain unclear and warrant further investigation Scalability: Training a separate model per sulcus simplifies the task but limits scalability and practical deployment, and does not exploit inter-sulcal relationships. It would be valuable to discuss whether future approaches could integrate all sulci within a unified framework, for example via multi-class segmentation. Task formulation: The problem is reduced to independent binary segmentation per sulcus, avoiding true multi-label labeling complexity. Comparison: Limited comparison to broader interactive segmentation methods beyond simple distance-based signals. Metrics: Dice score is appropriate but insufficient for small structures and interactive settings, as it does not capture boundary accuracy or user effort. Relevance: The problem of labeling small and variable sulci is relevant for certain neuroimaging studies, however, its broader impact on downstream neuroscience or clinical applications is not fully discussed or clearly demonstrated.

  • 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
    • Acronyms should be defined at first use. For example, ADT is introduced before being defined as angular distance transform, and WGDT should also be clearly expanded upon first appearance.
    • The separation between training and testing stages could be clearer, particularly regarding how user clicks are simulated during training versus evaluation. Validation:
    • It is unclear how the ground truth sulcal labels were generated and validated. Were expert neuroanatomists involved? Was inter-rater variability assessed?
    • More detail is needed on the validation protocol and annotation reliability. Methodological clarity:
    • It would be helpful to clarify whether the eikonal-based WGDT introduces significant computational overhead per interaction.
    • The assumption of correct user clicks should be discussed more explicitly as a limitation. For future work, I would recommend:
    • Evaluation with real user interactions to assess robustness and usability.
    • Exploration of unified multi-sulcus models or shared architectures instead of per-sulcus training.
    • Validation on broader populations (e.g., different ages, pathologies).
    • Inclusion of additional metrics (e.g., boundary accuracy, interaction efficiency).
    • Comparison to more recent interactive segmentation approaches beyond distance-based encodings.
  • 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?

    While the paper presents a novel and technically sound contribution with clear improvements over baseline methods, the evaluation relies on idealized simulated interactions, and important aspects such as runtime, robustness, scalability, and generalization are not sufficiently addressed. The work is interesting and relevant, but additional validation and clarification would strengthen its 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 #3

  • Please describe the contribution of the paper

    The authors propose an interactive spherical CNN-based framework for labeling cortical sulci, focusing specifically on the highly variable and anatomically challenging small sulci of the lateral prefrontal cortex. They encode interactions by solving the eikonal equation on the spherical surface, using a propagation speed function derived from the mean curvature of the white-matter surface

  • 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.

    Highly practical and simple method that also uses solid and appropriate mathematical techniques. Well written paper. Addresses a practical problem without getting overtly mathematical.

  • 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.
    • Human rater labeled ground truth sulci are lacking. Given that this is tedious to generate, a small-scale study would be helpful.
    • Effect of parameters on the performance needs to be evaluated with an ablation study.
  • 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.

  • 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?

    I think we need these types of papers that use just the right amount of math for addressing practical problems.

  • 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 all the reviewers for their constructive feedback and positive assessment of our method, highlighting a considerable advance over previous approaches (R1, R2), well-motivated and appropriate use of eikonal equation to encode user interactions (R2, R3) and soundness of the framework (R2). Please find our responses to the reviewers’ feedback below.

  • Guidance Signal / User Interaction (R1-W1, R2-W1,W2,W3,C2,C4,C5): As discussed in Sec. 2.3, the WGDT signal propagates guidance along sulci using mean curvature of the white matter surface (“curv”) in the speed function. Compared to “sulc” and “inflated.H”, “curv” better preserves narrow sulcal structures while reducing signal leakage into adjacent gyri. We used a simulated interaction during training and evaluation to maintain a controlled and reproducible setting, consistent with prior interactive segmentation studies (Sec. 2.2). While robustness to real-user or suboptimal interactions is an important future direction, our primary goal was to evaluate whether the proposed guidance signal better reflects expert annotator intent. Regarding computational cost, our framework is designed for real-time interaction. Both the SPHARM-Net backbone and the fast marching solver efficiently process high-resolution cortical surfaces (> 100k vertices), achieving end-to-end inference within well under a second per interaction in our implementation.
  • Problem Formulation / Architecture (R2-W5,W6,C5): We agree that extending the framework toward unified multi-sulcus interactive segmentation is an important direction. However, extending interactive refinement from a single sulcus to multiple anatomically variable structures is non-trivial, since user interactions occur across different sulci in arbitrary order and require simultaneous refinement of multiple targets. Therefore, as described in Sec. 2.1, we trained separate models for individual sulci to provide a controlled setting for evaluating the proposed WGDT guidance signal across distinct sulcal morphologies. Although a full multi-class framework remains challenging, we agree that integrating morphologically similar sulci within a shared model would be a meaningful intermediate direction.
  • Evaluation Protocol (R2-W7,W8,C3,C5,R3-W1,W2): The ground-truth sulcal labels were manually identified on individual cortical surfaces by multiple trained raters and subsequently reviewed and finalized by a neuroanatomist. Due to the labor-intensive nature of fine-grained sulcal labeling, constructing large-scale datasets remains challenging, as discussed in Sec. 1.We appreciate the suggestions regarding parameter analysis and evaluation metrics. For WGDT, multiple settings of the modulation parameter k were explored (Fig. 6), while the maximum propagation distance was fixed empirically. We agree that a more systematic analysis of propagation-related parameters, as well as additional evaluation metrics beyond Dice score (e.g., boundary-aware or interaction-efficiency measures), would be important future directions.
  • Scope / Generalization (R2-W4,W9,C5): Fine-grained sulcal labeling is becoming increasingly important as neuroimaging studies move beyond coarse parcellations to capture subtle morphological variability that differs across individuals, making fully automatic approaches less reliable and motivating interactive refinement (Sec. 1). Our experiments assume well-reconstructed surfaces and healthy participants. In pathological or excessively noisy cases, manual labeling itself becomes difficult and the task itself might be anatomically ill-posed. While such extreme conditions may affect curvature-based propagation, the WGDT signal remains effective within our experimental setting.
  • Typos / Minor Issues (R2-C1): We will carefully review the manuscript for minor issues such as typos, acronym definitions, and notation consistency. We will do our best to address these aspects thoroughly to ensure a clear and unambiguous presentation.



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.

    This submission proposes an interactive labeling method on spherical represerntation. After reweighting review content, the review feedback is found constructive and converges on methodological novelty of encoding user interaction via curvature driven diffusion, but also on major concerns on experimental design (simulated clicks, dataset generalizability). The authors are invited to clarify these concerns. Despite the necessary clarification, the submission in its core essence is found of sufficient novelty for recommending acceptance.



back to top