List of Papers Browse by Subject Areas Author List
Abstract
A generative model for the mesh geometry of intracranial aneurysms (IA) is crucial for training networks to predict blood flow forces in real time, which is a key factor affecting disease progression. This need is necessitated by the absence of a large IA image datasets. Existing shape generation methods struggle to capture realistic IA features and ignore the relationship between IA pouches and parent vessels, limiting physiological realism and their generation cannot be controlled to have specific morphological measurements. We propose AneuG, a two-stage Variational Autoencoder (VAE)-based IA mesh generator. In the first stage, AneuG generates low-dimensional Graph Harmonic Deformation (GHD) tokens to encode and reconstruct aneurysm pouch shapes. GHD enables more accurate shape encoding than alternatives. In the second stage, AneuG generates parent vessels conditioned on GHD tokens, by generating vascular centerline and propagating the cross-section. IA shape generation can be conditioned on specific clinically relevant shape measurements, enabling controlled studies on how morphological variations impact flow behaviors. Additional, our novel Morphing Energy Alignment constraint and Morphological Marker Calculator improve generation fidelity and controllability. Source code and implementation details are available at https://github.com/anonymousaneug/AneuG.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1474_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/1474_supp.zip
Link to the Code Repository
https://github.com/anonymousaneug/AneuG.git
Link to the Dataset(s)
AneuX morphology database: https://zenodo.org/records/6678442
BibTex
@InProceedings{DinWen_TwoStage_MICCAI2025,
author = { Ding, Wenhao and Ji, Kangjun and Castro, Simão and Luo, Yihao and Roi, Dylan and Yap, Choon Hwai},
title = { { Two-Stage Generative Model for Intracranial Aneurysm Meshes with Morphological Marker Conditioning } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15969},
month = {September},
page = {595 -- 604}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper introduces a conditional generation model for synthesizing meshes of intracranial aneurysms, including the aneurysm complex and parent vessels. It enables synthesis of aneurysm meshes that have specified clinical morphological measurements. It can be trained with a limited cohort of aneurysm shapes.
- 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 proposed method enables generation of intracranial aneurysm geometries conditioned on specific clinical morphological parameters, which is not feasible with alternative methods. This could be beneficial for fluid dynamic simulations that investigate rupture risk.
The centerline of the parent vessel is conditioned on the aneurysm geometry rather than the centerline being designed independently. The latter could generate unnatural examples.
- 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 is somewhat difficult to follow. Several of the losses are not explicitly defined. There is some confusing use of variables (Fourier mode coefficients phi are defined near Eq 6 following their first appearance in Eq 2; lowercase and uppercase psi are used without distinction; theta and z are not explicitly defined). The benefits of using the graph harmonic deform method are not clearly explained. Although the parent vessel centerlines are conditioned on aneurysm morphology, it seems from Sec 2.4 that they are nonetheless separately meshed and merged in a manner that is not entirely clear.
The evaluation approach is not entirely clear. The description does not specifically state how samples are generated with PCA and the diffusion shape generator for comparison. Units of the evaluation metrics are not given (Tables 1 and 2) and there is no color legend in Figure 2. While it is interesting that the proposed method can generate geometries by varying only specific shape features, it is not obvious how to interpret the wall shear stress maps in the generated models of Fig. 2G – a more descriptive caption may help. The flow simulation is not specified.
- 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
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 — 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?
The study is well motivated, however the presentation of the method and evaluation experiments would benefit from greater clarity.
- Reviewer confidence
Somewhat confident (2)
- [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.
Overall, this is an interesting framework for synthesizing realistic 3D meshes of intracranial aneurysms and their parent vessels, which has potential for clinical application. However, the rebuttal indicates that a number of updates would need to be made to the paper in terms of updating notation, describing methods details with more citations, and revising content such as the abstract. For this reason, the paper may not yet be ready for acceptance.
Review #2
- Please describe the contribution of the paper
This paper proposes a two-stage generative model for creating realistic 3D models of intracranial aneurysms, conditioned on specific morphological parameters. The first stage of AneuG focuses on generating the aneurysm complex, utilizing a Variational Autoencoder and Graph Harmonic Deformation tokens for shape encoding. The second stage generates the parent vessels, employing another VAE and Fourier tokens for shape representation. A key contribution of AneuG is the incorporation of a Morphological Marker Calculator, which allows for the control of the generated shapes based on clinically relevant parameters such as aspect ratio, neck width, lobulation index, and volume. This feature enables the generation of aneurysm shapes with specific clinical characteristics, which is crucial for medical applications. The authors evaluate their model using a dataset of 116 aneurysms located at the middle cerebral artery bifurcation, extracted from the AneuX dataset. They compare AneuG against a PCA statistical shape model and a deep learning Diffusion shape generator, demonstrating superior performance in terms of shape fidelity and realism. The paper presents a thorough evaluation of the model’s performance. The results indicate that AneuG is capable of generating realistic aneurysm geometries, outperforming the baseline methods. The authors also conduct ablation studies to assess the contribution of the Morphing Energy Alignment constraint and the Morphological Marker Calculator. The paper concludes by highlighting the potential of AneuG for medical research and clinical practice, particularly in simulating blood flow and assessing rupture risks. Overall, the paper presents an advancement in the field of aneurysm modeling, offering a novel approach to generating realistic and clinically relevant aneurysm geometries.
- 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 core strength lies in the novel two-stage generative model, AneuG, which effectively addresses the challenge of creating realistic 3D aneurysm models.
- The potential of AneuG to advance the field of medical image analysis and to be useful for researchers and clinicians working with aneurysm data is significant. The ability to generate realistic aneurysm geometries is crucial for simulating blood flow and assessing rupture risks, which has important implications for medical research and clinical practice.
- The inclusion of both realted quantitative metrics and a qualitative assessment by a neuroradiologist strengthens the evaluation.
- 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 suggest revising the abstract to better highlight the main contributions of this paper.
- The paper compares AneuG to only two baseline methods. I suggest providing more comparison of shape generation methods.
- There are bolded errors in Table 1, but the meaning of the bolding is not specified.
- The paper lacks a thorough discussion of the model’s limitations and potential failure cases.
- Please provide the variance (std) results in the experimental findings.
- Please explain why the results generated under different conditions in Table 2 show such significant differences.
- 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.
- 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 potential of AneuG to advance the field of medical image analysis and to be useful for researchers and clinicians working with aneurysm data is significant. The ability to generate realistic aneurysm geometries is crucial for simulating blood flow and assessing rupture risks, which has important implications for medical research and clinical practice.
- Reviewer confidence
Somewhat confident (2)
- [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.
My concern has been resolved.
Review #3
- Please describe the contribution of the paper
Authors proposed a new conditional generation model called AneuG that can generate realistic 3D models of intracranial aneurysms. These models are important for medical studies and simulations. This model has two-stage architecture for synthesizing realistic 3D meshes of intracranial aneurysms and their connected parent vessels. The first stage is an aneurysm shape generation module. It creates the main bulge and nearby vessel parts using a compact shape encoding method. The second stage is a parent vessel generation part. It adds the rest of the blood vessels in a way that smoothly connects to the aneurysm. The proposed model can also control the shape based on clinically important measurements; aspect ratio (AR) – height vs. neck size, neck width (NW), and lobulation index (LI). Both stages use a variational autoencoder (VAE), which helps it learn how to generate new 3D shapes from a small dataset. Graph Harmonic Deform (GHD) helps the model better understand aneurysm shapes. It also adds new constraints to make the shapes more realistic and consistent with clinical 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 introduces a new conditional generative framework for 3D mesh synthesis of intracranial aneurysms and their parent vessels. The two-stage design is novel in its explicit modeling of the anatomical relationship between the aneurysm and its vascular context. A significant of this work is its ability to control the generated shapes using clinically relevant morphological parameters. This makes the model highly valuable for clinical research applications such as shape-based risk assessment or targeted simulation studies. The use of Graph Harmonic Deformation (GHD) for shape representation is an effective alternative to traditional PCA or voxel-based methods. GHD enables the model to better preserve local geometric features, leading to smoother and more anatomically realistic meshes. Inclusion of comprehensive quantitative comparisons and clinical scenario validation further strengthens the impact of credibility of the proposed approach.
- 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 is trained and evaluated on a relatively small dataset of 116 aneurysms, all located at the middle cerebral artery bifurcation. This narrow anatomical focus may limit the generalizability of the model to aneurysms in other vascular territories or from more diverse patient populations. This paper does not include external testing on unseen datasets, nor does it assess whether the generated meshes are acceptable to clinicians beyond a single neuroradiologist ranking. The overall architecture is novel, but adaptations of techniques already present in prior work. It would be valuable to include a runtime and computational cost analysis.
- 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
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?
This paper presents a technically innovative generative framework for synthesizing realistic 3D meshes of intracranial aneurysms and their parent vessels. The proposed model is the first to explicitly model the joint distribution of aneurysm pouches and connecting parent vessels, which is a notable contribution in the context of anatomical shape generation for medical application. A key strength is the ability to control generated shapes using clinically meaningful morphological markers, enabling applications in shape-based risk assessment and training data augmentation. The integration of Graph Harmonic Deformation (GHD) for mesh encoding is effective in preserving local geometric details. The paper includes both quantitative metrics and clinical use case demonstrations, which support the model’s practical relevance for clinical research.
- Reviewer confidence
Somewhat confident (2)
- [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 reviewers and Chair for their efforts and guidance. Our GitHub link is in the abstract, we will add it to the main text. Reviewer 1
- How are two meshes merged? This is done in 3 steps: I: On the aneurysm complex, at the connections to the 3 vessels, short segments with fixed centreline length are clipped to create vascular openings (Fig 1A). We slice the centrelines by the same length when preparing stage II ground truth. II: At centreline points from Stage II, normals to centerline are used to define cross-sectional planes. The nodes that trace the vascular opening from (I) are projected onto these planes, forming tubular mesh nodes, which are algorithmically connected to form a triangular mesh. III: Stage I & II meshes are glued by inserting a loop of additional triangles at their interface gap. These details are given in the code. We will add this to the paper.
- Some losses not explicitly defined We will clarify definitions. In Sec. 2.1, L_cond is an MSE loss between the output of the MMC and ground truth. L_tReg is the MSE loss between vascular centerline tangents and those from the aneurysm complex (red arrows in Fig. 1a).
- Confusing variables, lack of units and legend We will define phi at Eq 2, change Psi in Eq 3 to lowercase to match Eq 2. We will clarify that in Eq 6 z is the VAE latent vector, and theta are GHD shape encoding tokens from Eq 1, used to condition Stage II. We will carefully vet other equations as well. Units to CD will be added to tables; other metrics are unitless [15]. Flow Colour legends (0-25Pa) will be added.
- Benefits of GHD Benefits of GHD is previously demonstrated [16]. As surface Fouriers, it is naturally smooth without regularization, yet it can more closely capture geometric details than other methods (coordinates-based mesh morphing, VAE, and PCA shape models). We will clarify this in the paper.
- PCA and Diffusion Methods For PCA: The first 15 modes of node coordinates (>99% variance) is used, coefficients are randomly drawn from the normal distributions of each mode. For Diffusion: we replicated the Michelangela method [15] but without image/text conditioning. As in Fig. 2 of [15], occupancy fields (from DVS in [16]) are fed into SITA-VAE. Details can be found in codes. We will add explanations to the paper.
- Flow Simulations: It’s done with ANSYS, we’ll add a citation and discuss results briefly: low, oscillatory WSS with big spatial variations is considered adverse, but studies are still ongoing to understand them [3]. Reviewer 2
- Revising abstract and tables We’ll clarify the overall contribution, add MMA and MMC strategies in the abstract, and clarify that bold metrics represent the best performance among different methods in Table 1, or different condition setup in Table 2, and provide standard deviations of metrics.
- Add more baselines methods for comparison We agree this is good to have, but we are not allowed to add new experimental results.
- Limitation We will add this: (1) conditional generation has limited accuracy, and lower quality due to small dataset size; (2) dataset of 116 is small; (3) a narrow anatomical focus on bifurcation IA, (4) only 1 radiologist providing feedback.
- Table 2: why different conditions cause big differences We’ll add explanations: with a small dataset, having more condition causes poorer generation, as the network doesn’t have enough examples of varying extent of each condition to learn from. Reviewer 3
- Not enough topologies and datasets We agree the focus on MCA bifurcation IA is narrow, but it has a high prevalence (43%) and we wanted to show that we can address a challenging topology. Our pipeline can be applied to other topologies. We also agree that more and larger datasets for further training and testing will strengthen the work, and will add limitations text.
- Computational cost We will add this. Preprocessing takes 30 min/shape, training <1 hour for 3 runs, and inference < 2 seconds on one RTX 3090 GPU.
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.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
While the paper presents an interesting framework, I agree with the reviewer that the promised updates seem to be substantial and unrealistic within the page limit in order to be ready for acceptance.