Authors

Huilin Yang, Roger Tam, Xiaoying Tang

Abstract

Reconstruction and visualization of cardiac structures play significant roles in computer-aided clinical practice as well as scientific research. With the advancement of medical imaging techniques, computing facilities, and deep learning models, automatically generating whole-heart meshes directly from medical imaging data becomes feasible and shows great potential. Existing works usually employ a point cloud metric, namely the Chamfer distance, as the optimization objective when reconstructing the whole-heart meshes, which nevertheless does not take the cardiac topology into consideration. Here, we propose a novel currents-represented surface loss to optimize the reconstructed meshes’ topology. Due to currents’s favorable property of encoding the topology of a whole surface, our proposed pipeline delivers whole-heart reconstruction results with correct topology and comparable or even higher accuracy.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43987-2_11

SharedIt: https://rdcu.be/dnwJt

Link to the code repository

N/A

Link to the dataset(s)

N/A

Reviews

Review #1

Please describe the contribution of the paper

This paper proposes a framework for automatically reconstructing meshes of cardiac structures. Firstly, a volume encoder-decoder is used to extract voxel features. Then, three deformation blocks are progressively applied to deform the initial ball-like meshes to fit the cardiac structures. To preserve the topology, a currents-represented surface loss is proposed.
Please list the main strengths of the paper; you should write about 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 is well-motivated, and the main contribution of the proposed methods appears to be the currents-represented surface loss, which better preserves the topology of the cardiac structure.
Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
1. Table 2 only presents three metrics, and it would be helpful if the authors could provide the quantitative results for all the metrics measured.
2. The results in Table 2 show that the performance on LV is better without L_s. It would be helpful if the authors could provide a reasonable explanation for this.
3.Since EMD (earth mover distance) loss is a commonly used loss in 3D shape reconstruction, it would be helpful if the authors could include some comparisons with it.
Please rate the clarity and organization of this paper

Very Good
Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

The results should be reproducible.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

It would be beneficial if the authors could conduct more experiments as suggested in the ‘Weaknesses’ section to further validate the proposed method.

Minor issues: The abbreviation ‘SI’ used in section 3 (Evaluation metrics) is not consistent with ‘IS’ used in Table 1.
Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

4
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

To summarize, I believe the proposed method is novel and well-motivated. However, I find the experimental results insufficient. I am open to changing my rating if the authors provide satisfactory results.
Reviewer confidence

Somewhat confident
[Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

N/A
[Post rebuttal] Please justify your decision

N/A

Review #3

Please describe the contribution of the paper

This paper proposes a novel approach for reconstructing cardiac structures from medical imaging data. This method uses currents-represented surface loss to optimize the topology of the reconstructed meshes, resulting in accurate whole-heart reconstructions.
Please list the main strengths of the paper; you should write about 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. The surface loss is based on a well-established mathematical concept, the computable norm on currents, which has been previously used in diffeomorphic surface registration. This seems to provide a theoretical foundation for the proposed approach. 2.The surface loss has extensive applicability in shape analysis and disease diagnosis, as noted in previous research. This suggests that the proposed approach could be useful in a range of medical applications beyond the specific case of whole-heart mesh reconstruction
Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

Based on the experiment section, it appears that there may be issues with the train-test configuration of the dataset, according to its limited size.
Please rate the clarity and organization of this paper

Good
Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

It seems OK
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

see the weakness section
Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

5
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

interesting design with topology
Reviewer confidence

Confident but not absolutely certain
[Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

N/A
[Post rebuttal] Please justify your decision

N/A

Review #2

Please describe the contribution of the paper

The authors propose a method for automatically generating whole-heart meshes directly from medical images. Their whole-heart reconstruction method optimizes the reconstructed heart topology by employing a new loss function, i.e., the currents-represented surface loss. It leads to accurate representation of the whole-heart topology.
Please list the main strengths of the paper; you should write about 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 main strength of the paper is the novelty of the method in that a new surface loss function is used which takes the cardiac topology into consideration. Hence the whole-heart meshes that the deep learning method reconstructs have accurate surface topology. Whereas previous methods employed the Chamfer loss function which was a point cloud-based metric which did not consider the topology.
Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

Runtimes were not included for the proposed method or the state-of-the-art methods. Hence it is not clear whether the proposed method is practical.

In addition, the paper did not discuss enough of the scientific computing details. For example, did the meshes ever tangled when they were being deformed onto the whole-heart geometry? If so, how was this handled? If not, is there a theoretical guarantee that they will not become tangled?

Finally, the writing of the paper needs to be improved. The writing does not strike the right balance between providing the motivation and the details on the mathematics, algorithmic formulation, and results. The paper is too detailed without including enough expository information. This makes it somewhat difficult to read in places.
Please rate the clarity and organization of this paper

Good
Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

The authors have provided most of the details that are needed for the paper to be reproducible. However, they are split between the paper and the supplementary material in somewhat of an odd manner. For example, the paper discusses the loss functions but not the optimization. However, deep learning cannot be done without the minimization of the loss function by an optimization method. Etcetera.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html
Please see the Strengths and Weaknesses noted above.

In addition, I have several suggestions for revisions which the authors should address prior to their paper being accepted for publication:
1. Abstract - “reconstructed meshes topology” should be “reconstructed mesh topology”.
2. The references that appear in a citation should be cited in order. For example, [1, 19, 12, 13, 14, 3, 4] should be [1, 3, 4, 12, 13, 14, 19].
3. Does the currents surface loss computation that is being used for the deep learning process represent a diffeomorphism of the surface mesh?
4. How is the ground truth mesh defined? Is this a ground truth surface which was obtained by manual segmentation and then a surface mesh was generated on it?
5. What do you mean by three deformation blocks? Do you mean three deformation stages?
6. The font that was used in FIgure 1 is too small. It is difficult to read. Please enlarge it.
7. Page 4 - “quite accuracy” should be “quite accurate”.
8. A smaller font size should not be employed for equation (4). Instead, the equation needs to be split over more lines.
9. What was done with the intersected mesh facets that were detected by Tetgen? Are they removed from the final mesh?
10. More description is needed of the deep learning process and terms.
11. What optimization method is used for the training? No details have been included regarding the choice of optimization method or the relevant parameters in the paper itself. Such details should not be relegated to the supplementary material section.
12. Does the optimization method identify a local minimum? Or a global minimum?
13. More details are needed in regards to the three deformations that lead up to the whole-heart meshes.
14. The text in Table 2 is too small.
15. Reference 15 - All proper names should be capitalized.
16. Small font should not be used for equations (6) through (9).
Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

5
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The proposed method to automatically generate topologically-accurate whole-heart meshes from cardiac medical images is an exciting development. The surface loss function which incorporates the surface topology of the heart is novel. However, the paper is currently too detailed and is not well-organized. The paper would need to be rewritten to include runtime details and more motivation/explanation and discussion on the implications of the results.
Reviewer confidence

Very confident
[Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

N/A
[Post rebuttal] Please justify your decision

N/A

Primary Meta-Review

Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. 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.

All reviewers agreed on the novelty aspect of using the currents-based representation for surface loss. One reviewer felt that the writing of the paper could be improved.

It was remarked that in addition to showing the Dice coefficient, Jacquard index, and Hausdorff distance, the average symmetric surface distance (ASSD), and the self-intersection (SI) metrics should also be displayed to get a better idea of the method. The first three metrics are general and will indeed measure the reconstruction accuracy. However the self-intersection metric is better suited to measure the consistency in topology, especially as this is the main point of the paper. One reviewer pointed the limitation in dataset size.

It is expected that the control of the scale parameter \sigma will change the neighborhood properties (from global to local and vice-versa). Although by itself, this is independent of topology. The authors should explain in detail why this approach of using currents-representation of surfaces is better suited for topological correctness. One reason is that it is independent of explicit parameterization as distances are only measured in the distributional sense. This property in addition to the scale parameter is largely at work here. While readers in this field will understand the theoretical aspects of the paper, the authors could explain this more clearly and provide further justifications for readability purposes.

Reviewers were also curious about the explanation of inclusion of the loss term L_s on the performance on the left ventricle as well as made a comment that the results could also show comparisons with the Earth Mover’s Distance. The authors should also comment about self-intersections (one of the reasons why IS should be displayed) and whether the meshes got tangled during deformation.

The authors should carefully look at the reviewers comments while preparing the rebuttal.

Author Feedback

N/A

back to top

Whole-Heart Reconstruction with Explicit Topology Integrated Learning