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

Abstract

Dental prosthesis is important in designing artificial replacements to restore the function and appearance of teeth. However, designing a patient-specific dental prosthesis is still labor-intensive and depends on dental professionals with knowledge of oral anatomy and their experience. Also, the initial tooth template for designing dental crowns is not personalized. In this paper, we propose a novel point-to-mesh generation transformer (DCrownFormer) to directly and efficiently generate dental crown meshes from point inputs of 3D scans of antagonist and preparation teeth. Specifically, to learn morphological relationships between a point input and generated points of a dental crown, we introduce a morphology-aware cross-attention module (MCAM) in a transformer decoder and curvature-penalty loss (CPL). Furthermore, we adopt Differentiable Poisson surface reconstruction for mesh reconstruction from generated points and normals of a dental crown by directly optimizing an indicator function using mesh reconstruction loss (MRL). Experimental results demonstrate the superiority of DCrwonFormer compared with other methods, by improving morphological details of occlusal surfaces such as dental grooves and cusps. We further validate the effectiveness of MCAM, MRL, and significant benefits of CPL through ablation studies. The code is available at https://github.com/suyang93/DCrownFormer/.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0638_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/0638_supp.pdf

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Yan_DCrownFormer_MICCAI2024,
        author = { Yang, Su and Han, Jiyong and Lim, Sang-Heon and Yoo, Ji-Yong and Kim, SuJeong and Song, Dahyun and Kim, Sunjung and Kim, Jun-Min and Yi, Won-Jin},
        title = { { DCrownFormer: Morphology-aware Point-to-Mesh Generation Transformer for Dental Crown Prosthesis from 3D Scan Data of Antagonist and Preparation Teeth } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15006},
        month = {October},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes a point-to-mesh generation transformer (DCrownFormer) to directly and efficiently generate dental crown meshes from point inputs of 3D scans of antagonist and preparation teeth.

  • 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. This paper proposes a point-to-mesh generation transformer that can capture geometric local-global features of point inputs at the transformer encoder and directly reconstruct fine details of a dental crown mesh at the transformer decoder.

    2. A morphology aware cross-attention module was introduced in a transformer decoder, which can capture the morphological relationships between a point input and generated points of a dental crown.

    3. A loss function combining curvature-penalty loss and mesh reconstruction loss is proposed.

  • 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. The ablation experiment was incomplete and did not reflect the improvement effect of MRL and CPL over baseline.

    2. There are gaps in the paper’s content in the Section of Evaluation Metrics, making it difficult to understand.

    3. There are challenges to the correspondence of various methods with Table 1. and Fig. 3 In the Section of Comparison with Other Methods, which is not conducive to the reader’s understanding.

  • 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    It is recommended that the dataset was open-sourced to improve the reproducibility of the proposed method.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    1. In the ablation study, this paper should add experiments that demonstrate the effect of MRL and CPL on baseline improvement.

    2. The missing content in the section of evaluation metrics should be improved.

    3. It is recommended to add references to the methods in Table 1.and Fig. 3, corresponding to the content of the Section of Comparison with Other Methods, to help readers understand.

    4. Why is the proposed method worse than the other methods in the Normal Consistency metric in Tables 1 and 2, and does it indicate limitations?

    5. Why is SAP included in all comparison methods in Tables 1’s comparison experiment, and why is SAP removal not included in any comparisons?

    6. It is recommended that the dataset was open-sourced to improve the reproducibility of the proposed method.
  • 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

    Weak Reject — could be rejected, dependent on rebuttal (3)

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

    Although this paper presented an interesting method, but there are missing content and description is not clear, and the comparative experiment is not comprehensive, which need to be improved.

  • Reviewer confidence

    Very confident (4)

  • [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

    This study introduces DCrownFormer, a deep learning model that leverages a point-to-mesh generation transformer to design patient-specific dental prostheses directly from 3D scans. This approach efficiently generates detailed dental crown meshes from point cloud inputs of antagonist and preparation teeth. The DCrownFormer significantly enhances the morphological accuracy of occlusal surfaces, offering a novel solution for creating highly customized and precise dental crowns.

  • 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 incorporation of a Morphology-aware Cross-Attention Module (MCAM) within a coarse-to-fine framework enables precise generation of the dental prosthesis geometry, tailoring it effectively to individual patient needs.
    2. The implementation of a Curvature-Penalty Loss (CPL) effectively constrains and enhances crucial morphological features such as dental grooves and cusps, which are essential for both the functionality and aesthetic quality of dental crowns.
    3. The proposed method adeptly utilizes both point data and normal information from 3D scans, leading to highly accurate outcomes.
  • 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.

    This study, exhibits some weaknesses that could affect its comprehensibility and broader application. Paper lacks sufficient explanation for general readers on key processes, such as the expansion process of g1 following the MAP. The study introduces IMS, described as potentially a single Multi-Layer Perceptron (MLP) layer, yet it is designated with a specific name. Clarification on why this naming was chosen would help differentiate the feature or elucidate its unique role in the model, if any. The term ‘indicator grid’ is used without a clear definition, leaving readers uncertain about its meaning and significance in the context of the study. Such terminological clarity is essential for a broader audience.

    1. Another limitation is the discussion on the generalizability of the proposed method. The paper focuses predominantly on dental crowns, with limited exploration of how this method could be adapted for other types of dental prostheses.

    2. ‘Future Works’ section of the paper is notably vague, lacking detailed plans or strategies for addressing the potential challenges or limitations identified in the current study.

  • Please rate the clarity and organization of this paper

    Excellent

  • 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    n/a

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    The current study should explicitly state its limitations, anticipated challenges in future research, and the proposed solutions. This detailed disclosure is crucial for guiding subsequent investigations effectively.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

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

    This study significantly advances the technology for designing dental crowns. The novel DCrownFormer method uses deep learning to create dental crowns from 3D scans quickly and precisely. This approach improves how crowns fit, look, and function, potentially making dental treatments better for patients. The results demonstrate superior performance compared to existing methods. Additionally, an ablation study validates the effectiveness of incorporating the Morphology-aware Cross-Attention Module (MCAM) and Curvature-Penalty Loss (CPL), proving critical to DCrownFormer’s success. The manuscript is well-structured and informative, providing sufficient methodological details to ensure reproducibility. Only minor revisions are needed to clarify specific points.

  • Reviewer confidence

    Very confident (4)

  • [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

    The authors propose a transformer based method called DCrownFormer to generate dental crowns meshed from the point cloud of the preparation tooth and antagonists. The authors introduce a novel loss called curvature penalty loss which is used to constrain the shape of the dental crown in terms of the grooves and cusps. The authors also introduce a morphology aware cross attention model to assist the network in understanding dental shape, scale and occlusion better. The results demonstrate that the proposed method outperforms other existing methods.

  • 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. introduction of a novel loss
    2. introduction of a novel module to assist the network in understanding the dental shape better as that is the key to dental crown generation and its aesthetics.
  • 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.

    I think the paper is well written.

  • Please rate the clarity and organization of this paper

    Very 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    The TopNet + SAM method seems to perform better on the metric normal consistency. Can you comment on this?

  • 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

    Strong Accept — must be accepted due to excellence (6)

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

    I think the proposed method is a novel point to mesh generation method for dental crowns.

  • Reviewer confidence

    Very confident (4)

  • [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



Author Feedback

Thanks to the reviewers for providing insightful and constructive comments.

[Reviewer #1-Q1] The TopNet + SAP method seems to perform better on the metric normal consistency. Can you comment on this? Author response: Thank you for your insightful and constructive comments. TopNet proposed a decoder following a hierarchical rooted tree to generate a structured point cloud. Although TopNet+SAP showed the highest performance on the metric normal consistency by hierarchically learning the overall structure of point clouds with normals of dental crowns, it had a limitation in learning the local details of dental grooves and cusps and the relationship to antagonist teeth, proximal teeth, and a margin line in this study. This results in the generated crown exhibiting a somewhat smoothed appearance. On the other hand, the DCrownFormer reconstructed the complex occlusal surface structure in more detail.

[Reviewer #3-Q1] The term ‘indicator grid’ is used without a clear definition, leaving readers uncertain about its meaning and significance in the context of the study. Such terminological clarity is essential for a broader audience. Author response: Thank you for your insightful and constructive comments. We use the Differentiable Poisson surface reconstruction method [19] to obtain an indicator grid, which implicitly determines the mesh surface through the Poisson equation [19] and can be converted to a mesh using Marching Cubes [23].

[Reviewer #3-Q2-3] Another limitation is the discussion on the generalizability of the proposed method. The paper focuses predominantly on dental crowns, with limited exploration of how this method could be adapted for other types of dental prostheses. ‘Future Works’ section of the paper is notably vague, lacking detailed plans or strategies for addressing the potential challenges or limitations identified in the current study. Author response: Thank you for your insightful and constructive comments. We fully agree with your comment that our study has limitations in generalizability for applying other types of dental prostheses. Dental prostheses are generally classified into five types, including dental implants, veneers, dentures, crowns, and bridges. Also, each type serves different purposes and comes with its own set of challenges and considerations. We carefully think that our study has the potential to be extended to future studies for the mesh generation of other types of dental prostheses.

[Reviewer #5-Q1] The ablation experiment was incomplete and did not reflect the improvement effect of MRL and CPL over baseline. Author response: Thank you for your comment. We evaluated the effectiveness of MRL and CPL in Table 2b and Fig. 3c, respectively. The baseline with MRL achieved higher performance than without MRL (Table 2b). Also, the proposed CPL (λ = 1.0) outperformed CDL (λ = 0.0) in terms of CD and SDE in Fig. 3c.

[Reviewer #5-Q2] There are gaps in the paper’s content in the Section of Evaluation Metrics, making it difficult to understand. Author response: We will improve the Evaluation Metrics in future work.

[Reviewer #5-Q3] There are challenges to the correspondence of various methods with Table 1. and Fig. 3 In the Section of Comparison with Other Methods, which is not conducive to the reader’s understanding. Author response: In Fig. 3, the color map denotes surface distance errors reflecting the SDE in Table 1, and the green and red indicate a low and high SDE, respectively. We will add generation results about points and indicator grids in future work.

[Reviewer #5-Q4] Why is SAP included in all comparison methods in Tables 1’s comparison experiment, and why is SAP removal not included in any comparisons? Author response: The purpose of our study is direct point-to-mesh generation using point completion networks combined with SAP, where SAP is used to mesh reconstruction from generated points and normals of a dental crown. Therefore, a comparison of SAP removal is not provided.



Meta-Review

Meta-review not available, early accepted paper.



back to top