Abstract

The automated 3D detection and segmentation of intracranial aneurysms is critical for clinical intervention, yet existing deep learning methods often struggle with the complex, tubular organization of cerebral vasculature. We propose EXPIGEO, a geometry-aware framework that bridges this gap by integrating explicit geometric reasoning with topological graph representation. Standard local features are easily confounded by the inherent curvilinearity of the vasculature, heavily limiting the effectiveness of conventional Graph Neural Networks (GNNs). EXPIGEO overcomes this by introducing an internal geometric exploration mechanism to characterize the vessel interior, enabling the robust extraction of skeletal topology and physiological descriptors. This explicit structural encoding successfully resolves topological ambiguities and unlocks GNN performance, establishing a new state-of-the-art benchmark for aneurysm detection on the IntrA dataset with 98.41% accuracy and an F1-score of 0.9589. Beyond raw performance, EXPIGEO provides clinical transparency: instance-level explainability confirms that predictions are driven by biologically relevant volumetric features, ensuring a reliable and interpretable diagnosis.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2646_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/mohamedaminelayachi/EXPIGEO

Link to the Dataset(s)

https://github.com/intra3d2019/IntrA

BibTex

@InProceedings{LayMoh_Exploiting_MICCAI2026,
        author = { Layachi, Mohamed Amine AND Nouri, Anass AND Eddarouich, Souad AND Autrusseau, Florent},
        title = { { Exploiting Interior Geometry for Intracranial Aneurysm Detection and Segmentation on 3D Point Clouds via GNNs } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16885},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes EXPIGEO, a geometry-aware framework for intracranial aneurysm detection and segmentation on 3D point clouds. The key idea is to explicitly model the interior vascular geometry using a ray-probing mechanism, enabling the extraction of structural representations, such as centerlines and radius fields. The authors design a set of geometric descriptors to be used as node features in a graph neural network for joint detection and segmentation. Experiments on the subset ofIntrA dataset demonstrate strong performance, particularly in aneurysm detection.

  • 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 paper addresses an important and clinically relevant problem with a clear motivation: standard point-based features are insufficient for modeling complex vascular geometries.

    2.The proposed interior geometry exploration and the resulting geometric descriptors are well designed, intuitive, and physiologically meaningful.

    3.The experimental results are convincing, with strong performance (especially in detection) and a well-executed ablation study demonstrating the necessity of the proposed features.  The inclusion of explainability analysis is a valuable addition, particularly for clinical applications where interpretability is important.

  • 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 overall novelty is somewhat incremental, as the contribution mainly lies in integrating geometric processing and feature engineering rather than introducing fundamentally new modeling approaches.

    2.The pipeline is relatively complex and involves multiple stages, which may introduce additional computational cost and potential error propagation.

    3.The evaluation is limited to a relatively small dataset(subset of IntrA), and additional validation would strengthen claims of generalization.  

  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

  • 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.The overall novelty of this paper is incremental.

    2.The experiment result is based on the subset of IntrA, so the generality is doubted.

  • Reviewer confidence

    Confident but not absolutely certain (3)

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

    Reject

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

    The authors’ response failed to address my comments, so I maintain my original score.



Review #2

  • Please describe the contribution of the paper

    The main contribution is a method to detect aneurysms from blood vessel meshes, leveraging mesh level methods

  • 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 authors address a relevant topic

  • 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 role and novelty of the proposed centerline extraction strategy are not fully convincing. Centerline extraction is a well-studied and largely solved problem, and it is unclear how the proposed approach improves over existing state-of-the-art methods. In particular, the need to reconstruct connectivity from ray probing and subsequent skeletonization is not sufficiently justified, and may introduce unnecessary complexity without clear benefits. -The definition and computation of the geometric sphericity feature are not clearly specified. It is unclear whether this measure is computed over the entire vascular surface, or restricted to the detected aneurysm region. This ambiguity makes it difficult to interpret its relevance and contribution to the model. -The overall purpose and role of the neural network within the framework are not clearly explained. The paper focuses on implementation details, but lacks a clear high-level description of what the model is intended to learn and how it contributes to the final task. Additionally, the choice of operating on the full surface representation is not sufficiently justified, especially given that parametric or centerline-based representations are commonly used in vascular modeling. A discussion or comparison would strengthen the motivation. -The formulation of the tasks is unclear. The paper refers to both detection and segmentation, but does not clearly define the distinction between them. It is also unclear how different classes (e.g., aneurysm, non-aneurysm, vessel) are modeled and evaluated. This lack of clarity makes it difficult to understand the learning objective and properly interpret the reported results. -The experimental validation could be strengthened. In particular, Figure 2 would benefit from a clearer comparison between the proposed method and the ground truth (difference between them and not each separate), as well as the inclusion of additional baseline methods. This would help better assess the qualitative and quantitative improvements of the approach.

  • 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 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

    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?

    i believe several issues need to be addressed for the paper to be accepted

  • 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.

    I believe there is real value in the work, and I thank authors for addressing reviewers concerns



Review #3

  • Please describe the contribution of the paper

    1.A novel and robust framework to extract centerline from tubular shape. 2.Construction of graph based on the extracted centerline locations and the surface points. 3.Extracting clinically relevant feature based on radius gradient.

  • 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.

    Solid methodology, concrete solution formulation and analysis, well presented paper. The performance metric is strong.

  • 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.Previous thinning algorithm such as [TC Lee,1994] were not mentioned or considered for testing.

  • 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.

  • 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?

    Well-written, solid centerline-extraction algorithm and have the potential to be widely applied to any tubular structure.

  • 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

We thank the reviewers for constructive feedback. We address the major issues raised by reviewers below.

Reviewers inquired about our pipeline’s novelty, the necessity of ray-probing over standard mesh algorithms, and comparisons to volumetric thinning like Lee et al., 1994 (R1.1, R2.1, R3.1). To clarify, our framework operates directly on discrete point clouds rather than connected meshes–a distinction we will emphasize in the revision. Adapting mesh-based or volumetric algorithms (e.g., Lee 1994) to sparse point clouds introduces massive memory overhead and noise sensitivity. Specifically, volumetric thinning is prone to generating spurious branches. Because our explainability analysis proves “blockage” is a crucial predictive feature, false spurs would artificially inflate false positives. Furthermore, iterative voxel peeling cannot capture the continuous radius field (also proven crucial by explainability) that our ray-probing natively extracts without surface reconstruction. Our novelty lies in bridging this explicit geometry with topological learning. This resolves the representation bottleneck of GNNs on sparse point clouds, bypassing coordinate variance to unlock robust performance for curvilinear structures.

Addressing concerns regarding multi-stage error propagation, task definitions, and computational cost (R1.2, R2.3, R2.4), we note that although our framework involves multiple stages, it operates entirely end-to-end during inference, preventing disjointed error propagation. The GNN performs binary node classification on the full surface point cloud. We explicitly group healthy vessels and non-aneurysmal bifurcations into a single negative class, isolating aneurysms as the positive target. We process the full surface because clinical planning requires exact 3D boundaries rather than simplified parametric models. This single node-level prediction inherently drives two distinct evaluation tasks: (1) Segmentation, assessing spatial precision at the point level, and (2) Detection, an instance-level task filtering localized candidates requiring >=10% IoU. We will update the text to define these tasks clearly. Computationally, ray-probing uses KD-Trees to bound extraction complexity to O(N log N) time and O(N) space. Average inference takes only 0.227 +/- 0.058s (CPU) for ray-probing per 512-point cloud and 13.1 +/- 2.8ms for the GNN (GPU).

Regarding the clarity of how geometric sphericity is computed (R2.2), Geometric Sphericity (phi) is a local descriptor extracted across the entire vascular bifurcation prior to downstream tasks. For each skeletal vertex v, we isolate its corresponding surface neighborhood (Omega_v), compute the cluster’s sphericity, and broadcast this value back to all points within Omega_v. This creates a continuous feature field that explicitly helps the model distinguish spherical aneurysms from tubular vessels.

In response to concerns about relying on a subset of the IntrA dataset and model generalization (R1.3), we note that IntrA (116 annotated aneurysmal bifurcation samples) remains the most standardized and rich 3D IA benchmark. Using it is essential for ensuring fair comparisons against baselines like EPT-Net. To maximize evaluation rigor, we enforce a strict 34-sample hold-out test split. Furthermore, while MICCAI rules preclude adding new external experiments during the rebuttal phase, our existing internal validation on two private clinical datasets has demonstrated similar generalization. This reinforces the robustness of our geometric priors and shows that our standardized protocol on IntrA sufficiently validates the framework.

To address requests for clearer visualizations and baselines (R2.5), we will update Fig 2 to overlay ground-truth boundaries onto network predictions, explicitly highlighting error margins. We will also add a qualitative comparison against a baseline model (a GNN using only XYZ coordinates) with an intuitive color scheme.




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.

    Mixed reviews with reasonable justification lead to rebttual. The authors should provide rebuttal in terms of experiment setttings, method novelty and other points in the reviews.

  • 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.

    There is 1 Reject and 1 Accept after rebuttal. R3 did not update the final recommendation but was initially strongly positive. The rebuttal clarified several concerns on the task definition, geometric features, and computation. Some concerns remain on novelty and generalization, but overall the method seems solid and useful, and the positive reviews are more convincing after rebuttal. I recommend acceptance.



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.

    The paper received mixed initial reviews, but the main concerns were largely related to clarification, positioning, and validation scope rather than fundamental issues with the core idea. After considering the rebuttal and reviewer updates, I recommend acceptance.

    The main strength of the work is its explicit exploitation of interior vascular geometry for point-cloud-based aneurysm analysis. The ray-probing mechanism and derived geometric descriptors are physiologically meaningful, and the results show strong detection performance, clear gains over coordinate/curvature-only GNN baselines, and useful explainability analysis.

    The rebuttal clarifies several key issues, including the motivation for point-cloud-specific geometry extraction, the role of the GNN, the detection/segmentation formulation, and the computation of geometric sphericity. For the camera-ready version, the authors should incorporate these clarifications, better position the method against classical centerline/thinning approaches, and more clearly discuss the limited external validation.



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.

    This paper addresses an important and clinically relevant problem, and the proposed EXPIGEO framework is reasonably well motivated, with an intuitive interior-geometry design and clinically meaningful geometric descriptors. The rebuttal also clarifies several key concerns, including the point-cloud setting, the task formulation, and the computation of the geometric features. Although the evaluation is still limited to a subset of IntrA, I believe the methodological idea and the explainability analysis make this a valuable contribution. Overall, I am inclined to accept this paper.



back to top