List of Papers Browse by Subject Areas Author List
Abstract
Accurate patient-specific tetrahedral cardiac meshes are essential for in-silico trials, yet common segmentation-then-modelling pipelines can blur thin-wall anatomy and offer limited cross-case correspondence. We propose HeartVolMesh, which lifts each template vertex to an anisotropic Gaussian kernel and uses a 3D CNN-GNN to predict per-vertex displacements and Cholesky-parameterized covariances from volumetric images. Training is guided by a covariance-aware negative log-likelihood loss with lightweight mesh regularization. For volumetric meshing, we warp a fixed tetrahedral template to the reconstructed surface via staged alignment, non-rigid registration, and deformation propagation, preserving connectivity and correspondence by construction, with resolution controlled by template density. Experiments show consistent gains over deformation-based baselines in surface mesh accuracy and volumetric mesh fidelity.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3675_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/ccmim/HeartVolMesh
https://fmlinks.github.io/projects/heartvolmesh/
Link to the Dataset(s)
N/A
BibTex
@InProceedings{LinFen_HeartVolMesh_MICCAI2026,
author = { Lin, Fengming AND Zakeri, Arezoo AND Dou, Haoran AND Zhou, Zherui AND Lan, Shaokun AND Duan, Jinming AND Frangi, Alejandro F.},
title = { { HeartVolMesh: Cardiac Volumetric Mesh Reconstruction via Covariance-Guided Graph Deformation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16889},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents “Heart VolMesh,” a comprehensive two-stage framework designed for the accurate reconstruction of patient-specific tetrahedral cardiac meshes from 3D volumetric images. Recognizing the limitations of traditional segmentation-then-meshing pipelines, the authors propose a novel approach. The first stage utilizes a covariance-guided 3D CNN-GNN to predict surface meshes, introducing an anisotropic Gaussian kernel representation to handle boundary ambiguities. The second stage employs a template-driven volume-to-surface registration strategy to warp a high-quality tetrahedral template to the reconstructed surface, thereby preserving internal connectivity and cross-case correspondence. The method is extensively evaluated on a large-scale, multi-center in-house dataset comprising 900 patients, demonstrating notable improvements in both surface accuracy and volumetric mesh fidelity over existing deformation-based baselines.
- 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.
-
Promoting deterministic vertices to anisotropic Gaussian kernels via covariance-guided graph deformation is a clever and logical adaptation for handling the spatially varying uncertainties inherent in medical imaging.
-
The manuscript is well-written, and the visual presentation (e. g. , the pipeline schematic and performance radar plots) is of high quality. Furthermore, validation on a large cohort of 900 patients demonstrates the robustness and clinical applicability of the proposed pipeline.
-
- 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 proposed two-stage framework, particularly the template-driven volumetric warping pipeline, shows great potential for standardizing cardiac digital twin generation. However, the specific implementation details of the non-rigid registration and the high-quality tetrahedral templates are crucial for reproducibility. It is strongly recommended that the authors open-source their codebase, pre-trained models, and the standard tetrahedral templates upon publication. Releasing these assets will significantly benefit the research community, allowing peers to readily adopt this pipeline for downstream FEM/CFD simulations and further advance the field.
- 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.
(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 well-motivated, technically sound, and clinically highly relevant framework for cardiac volumetric mesh reconstruction. The introduction of covariance-guided graph deformation to handle spatial uncertainty is an elegant design choice, and the extensive validation on a large cohort of 900 patients provides strong, convincing evidence of the method’s robustness and clinical utility. The manuscript is well-written, and the visual presentation is good.
- 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.
I appreciate the authors’ thorough and constructive response to all reviewers’ comments. The commitment to open-sourcing the entire repository (including the code, 24-part tetrahedral templates, and weights) effectively addresses the reproducibility concerns raised by myself and other reviewers.
Review #2
- Please describe the contribution of the paper
- The paper introduces the automatic creation of a volumetric tetrahedral mesh for five cardiac structures
- Introduction of covariance negative log likelyhood loss for vertext morphing
- Generation of a mesh by transformation of a template mesh to the predicted Surface. This is done per structure and then combined for a global deformation field that is applied to the template
- 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.
- Introduction of covariance based loss for graph deformation improves accuracy of the surface.
- The integration of several deformation fields into one global deformation field that is applied to deform the volumetric mesh template leads to anatomical plausible results while still using information of the different cardiac structures.
- 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.
- No evaluation on external/public dataset
- Performed test for statistical significance is not described e.g. what is the reference method? Are multiple tests performed and is it adjusted for multiple testing?
- Inaccuracy in Fig. 1 the initial template is most probably deformed by features from the latent space while the last graph deformation is performed using features from the highest resolution conv stage. Also the input data goes into the second stage of the CNN encoder
- 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.
- 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
- Fig. 1 and 3 are not referenced in the text. Fig. 4 is referenced but not discussed. What are the metrics normalized to?
- Probable mistake in Fig. 1 the initial template is most probably deformed by Features from the latent space while the last graph Deformation is performed using Features from the high Resolution conv stage. Also it Looks like the Input data goes into the second stage of the CNN encoder
- 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?
The methodology is sound but the methodological advancement is incremental. CNN feature extraction together with graph deformation already exists, the advancement here is the convariance guided NLL based loss. The template based volume creation is well done but the underlying non-ridigid Surface Registration used for deformation field creation is not new. Figure 2 needs to be corrected
- 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 paper presents “Heart VolMesh,” a framework for automatic reconstruction of topology-consistent, patient-specific tetrahedral cardiac meshes from 3D CTA volumes. The objective is to produce meshes with cross-case vertex correspondence accross a dataset. To this end, the authors introduce a Covariance-Guided Graph Deformation network that represents mesh vertices as anisotropic Gaussian kernels with learned covariance matrices, enabling the model to account for spatially varying geometric uncertainty. This is combined with a Template-Driven Volumetric Meshing that propagates surface deformation fields to a high-quality tetrahedral template, ensuring consistent vertex correspondence and simulation-ready mesh quality.
- 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.
S1 Innovative Representation: Promoting mesh vertices from points to anisotropic Gaussian kernels is a way to handle the spatially varying uncertainty and thin-wall ambiguities inherent in cardiac CTA imaging.
S2 Consistency & Correspondence: The template-driven warping approach successfully enforces cross-case correspondence, which is a significant advantage for population-level analyses and standardized downstream simulations.
S3 Comprehensive Validation: The paper provides robust quantitative evaluation across a large number of anatomical structures (24 parts), demonstrating that the framework scales well beyond simple ventricular models.
- 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.
W1 Limited motivation of the model: the 2 contributions are described but are not motivated. For instance, it is not clear at all the benefit of adding a covariance matrix per node. The notion of covariance and how this helps to better reconstruct the mesh should be explained.
W2 Algorithm complexity : While the surface reconstruction is end-to-end, the volumetric warping step (especially with multiple structure-specific deformation fields) seems to introduce a multi-stage process at inference. The paper could clarify the total runtime more explicitly.
W3 Limited details : The combination of the deformation fields from structures into a single field is not described in sufficient details (the gaussian blur step in particular),
W4 Hyperparameter Sensitivity: The model introduces several hyperparameters (e.g., $\beta$ for the loss function, weights for regularizers). The sensitivity of the results to these specific values is not deeply explored, which could impact reproducibility on different clinical datasets.
W5 Template Dependency: The final quality of the volumetric mesh is inherently linked to the topology and resolution of the chosen tetrahedral template. The paper could better discuss the limitations imposed by this fixed template structure, particularly for highly pathological cases that may deviate significantly from the template’s topology.
- 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.
- 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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper looks sound but it is not reproducible (code and data are private) and it is not well motivated. I am afraid will not bring any new tangible knowledge to the community.
- 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.
The fact that the authors will release their code is an important new fact.
Author Feedback
We thank the reviewers for their valuable feedback. We focus on the main concerns below.
Meta/R1/R2/R3 Reproducibility. We have prepared project homepage and github repo with two-stage code, configs, evaluation scripts, 24-part tetrahedral templates, and inference weight. No links are included to preserve anonymity and follow rebuttal rules; the release link will be added after acceptance.
Meta/R3W1 Method motivation. Our first contribution is covariance-guided graph deformation. Downstream simulation failure often arise from geometric error at anatomical junctions and thin-wall regions. Per-vertex covariance models spatially varying uncertainty and direction-dependent tolerance caused by partial volume, thin walls, and junction ambiguity, consistent with Reviewer1-S1.Ablations support this: LVMyo CD improves from 4.5/5.8 to 3.6 and RVMyo CD from 3.9/4.5 to 3.1.Our second contribution is volume-to-surface registration for volumetric meshing. In silico trials require volumetric meshes as inputs for EP/CFD/FEM solvers, while almost all image-to-mesh methods output only surfaces; heuristic tetrahedralization can lose cross-case/cross-phase correspondence and case-dependent connectivity. We use learned surface geometry to drive a dense deformation field that warps tetrahedral templates while preserving connectivity and vertex indices. The field can be sampled on compatible LV+aorta or whole-heart templates with different densities/elements, the method is not tied to one fixed template.
R2W1 External validation. The submitted evaluation already uses a large multi-centre cohort from multiple sources: 900 patients and 4004 NIfTI images, with a patient-level split to avoid leakage, R1 also recognized this large-scale validation. Public validation on ImageCAS/MMWHS will be important future work.
R2W2 Statistics. We used paired two-sided t-tests comparing Ours3p6 with each baseline/ablation on the same cases. Multiple tests were unadjusted, p<0.05 is descriptive, not family-wise corrected.
R2W3 Figures. The reviewer is correct: early deformation uses low-resolution/deeper features for global alignment, while final deformation uses higher-resolution/shallower features for local refinement. The arrows were simplified to avoid line crossings; this drawing error will be corrected.
R3W2 Complexity/runtime. In our implementation, each case produces 24-parts volumetric meshes with about 1M total vertices in <90 s on one NVIDIA A100 GPU. Stage 1 reconstructs the whole-heart covariance-guided surface mesh in one parallel network inference, not by per-part inference. Stage 2 performs structure-wise registration/warping sequentially; the learning stage is shared, while registration remains modular and parallelizable.
R3W3 Implementation details. After per-structure registration, LA/LVMyo/RA/RV/Aorta surface displacements are merged into one sparse set. Each grid node is assigned by nearest-neighbor KD-tree to build a 200^3 field, followed by Gaussian blur (sigma=5), cubic upsampling to 300^3, and another blur (sigma=10). Volume vertices are warped by cubic interpolation. CPD alpha/beta/lambda/max-iter=0.01/70/1/200; aorta uses ICP plus partial TPS. Full code will be released if accepted.
R3W4 Hyperparameters. Key settings are specified: 128^3 patches, beta=0.5, regularization weights 0.1, and 300^3 final field. The implementation further uses 3000 surface samples, covariance jitter 1e-5, and Cholesky floor 1e-3.These were selected by grid search on validation data; the same strategy can be used for other clinical datasets. Submitted ablations cover covariance NLL, field coherence/resolution, and mesh regularization.
R3W5 Template dependency. Cause we use deformation field, the method can use templates with different element types, densities, and cardiac structures. The method can capture topology-preserving pathological changes, e.g., myo thickness variation. We acknowledge that topology-changing anatomy cannot be generated.
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.
The paper presents a clinically relevant and well-validated framework for cardiac volumetric mesh reconstruction with strong large-cohort results, though its methodological contributions are somewhat incremental and reproducibility is limited. With two positive and one negative rating, the AC recommends proceeding to rebuttal before making a final decision.
- 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 authors have largely clarified the previously unclear points in their rebuttal, and the reviewers appear satisfied with the responses, with agreeing on acceptance. The AC therefore recommends 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 authors addressed the reproducibility concerns by committing to release the complete codebase, templates, and trained models upon publication. The reviewers reached a unanimous consensus to accept the paper following this clarification.
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.
The paper, after rebuttal, is sufficiently strong for acceptance. The work addresses an important problem in patient-specific cardiac volumetric mesh reconstruction and proposes a technically sound framework combining covariance-guided graph deformation with template-driven tetrahedral mesh warping. The validation on a large 900-patient cohort provides strong evidence of robustness and clinical relevance. The rebuttal also addressed a key reproducibility concern through the authors’ commitment to release the full repository, including code, templates, and model weights. Concerns about external validation, statistical testing details, and clearer motivation of the covariance formulation should be addressed in the final version.
