Abstract

Recent advances in deep learning have demonstrated remarkable potential in automating dental crown design. However, existing evaluation protocols predominantly rely on geometric similarity metrics (e.g., Chamfer Distance), which are clinically agnostic and fail to capture functional failures such as occlusal interference or open contacts. To bridge this gap, we present CrownBench, a physics-aware evaluation framework and benchmark that shifts the paradigm from geometric approximation to clinical validity. We propose a set of clinically-interpretable digital metrics, mathematically formulating implicit clinical experiences into explicit constraints: Feasibility Metrics (e.g., Occlusal/Proximal Feasibility Rates) for survival validation, and Quality Metrics (e.g., Occlusal Contact Distribution) for personalized quality assessment. Benchmarking state-of-the-art methods on CrownBench reveals a severe “Geometric Trap,” where geometrically accurate models still suffer from low clinical success rates. Furthermore, a rigorous blind user study confirms that our framework aligns significantly better with human expert judgments than traditional metrics. Our code and dataset are released at https://github.com/JiaminWoo/CrownBench.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1259_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/JiaminWoo/CrownBench

Link to the Dataset(s)

https://github.com/JiaminWoo/CrownBench

BibTex

@InProceedings{WuJia_From_MICCAI2026,
        author = { Wu, Jiamin AND Huang, Wendong AND Gu, Jiaying AND Hao, Jing AND Fu, Yuanfei AND Wang, Yilei AND Lai, Peichao AND Tsoi, James Kit Hon},
        title = { { From Geometry to Clinic: A Physics-Aware Evaluation Framework and Benchmark for Automated Dental Crown Design } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16893},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper introduces CrownBench, a clinically grounded benchmark for evaluating automated dental crown design beyond geometric similarity. It shows that the proposed physics-aware metrics better reflect clinical validity and expert judgment than conventional geometric metrics.

  • 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.

    Major strengths of the paper include its attempt to move beyond conventional geometry-based evaluation and to discuss clinically oriented evaluation metrics for dental crown design, which is a relevant and potentially useful direction for the field.

  • 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.

    Major weaknesses of the paper include its limited novelty, as the proposed metrics mainly repackage clinically intuitive criteria into an evaluation framework rather than introducing a fundamentally new methodology. More importantly, these metrics seem to offer limited practical guidance for improving crown design algorithms themselves, whereas the central challenge in this area is still methodological advancement in generation and optimization. In addition, the paper may overstate the distinctiveness of its evaluation setting. Recent studies have already begun to adopt multiple evaluation perspectives beyond simple geometric similarity, making the claimed gap less compelling than presented.

  • 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.

  • 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?

    The paper contains several noticeable inconsistencies and presentation errors, including mismatched dataset/test-set sizes, unclear metric definitions, and inconsistent numerical scales in Table 1.

  • Reviewer confidence

    Very confident (4)

  • [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.

    N/A



Review #2

  • Please describe the contribution of the paper

    The authors focus on automated crown design and propose a multi-dimensional evaluation framework that extends assessment from geometric similarity to clinical usability. The paper also includes extensive experiments to demonstrate the effectiveness and practical value of the proposed evaluation framework.

  • 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 introduce a set of clinically oriented evaluation metrics for automated crown design and validate, through extensive experiments, their effectiveness compared with traditional metrics such as Chamfer Distance.

  • 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.In Table 1, the ADF score of Human Design is reported as 100, which appears questionable and should be checked carefully. 2.In the Human-AI Alignment Validation section, the authors state that three senior dentists participated in the experiments, but the paper does not clearly describe how the dentists’ subjective ratings were collected, nor does it report inter-rater agreement. Providing such details and results would further strengthen the robustness of the experimental validation. 3.Although the proposed metrics can all be computed directly, this type of evaluation criterion may lack flexibility and may be difficult to adapt to real-world scenarios. Moreover, in Figure 2, the Spearman correlation for the overall dimension is only 0.52.The authors should further discuss: (1) whether these metrics could be better integrated in practice, for example through weighting schemes or other aggregation strategies; and (2) what advantages this metric-based approach has over methods based on model-encoded similarity.

  • 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.

  • 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?

    This paper addresses an important problem in automated crown design and proposes a clinically motivated evaluation framework beyond conventional geometric similarity metrics. The idea is meaningful and supported by extensive experiments, making the work potentially valuable. However, important concerns remain, including unclear details in the human evaluation, limited discussion of metric integration and practical robustness, and a possible error in Table 1.

  • 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.

    N/A



Review #3

  • Please describe the contribution of the paper
    • A set of quantitative metrics designed to evaluate quality of dental crown.
    • A rigorous blind user study on a subset of the 500 paired molar crown cases confirms that the proposed metrics aligns significantly better with human expert judgments than traditional metrics.
  • 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 framing of the problem space: existing evaluation protocols predominantly rely on geometric similarity metrics(e.g., Chamfer Distance), which are clinically agnostic and fail to capture functional failures such as occlusal interference or open contacts. Thus need better evaluation method.
    • The design of the metrics set is clinically-motivated and embeds domain knowledge into the feasibility metrics and quality metrics.
    • Comprehensive benchmarking of multiple generative models which in itself is a good resource for the research field, and meanwhile along with the correlation study from blind user study, also shows the capability of the proposed metrics to distinguish different methods which aligns with human expert judgement.
  • 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 main contributions are metric design and benchmarking, lacking in solid modeling or algorithmic advances
  • 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.

  • 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.

    (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 major strengths of the paper (see answer to question 6) are equally battling with the major weakness of the paper (see answer to question 7). Considering this paper is very well organized(solid motivation, clear mathematical formulas, well-designed experiments including benchmarking and blind user study, and nice listing of limitations and potentials), my overall feedback is a weak accept

  • 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 #4

  • Please describe the contribution of the paper

    The main contribution of this paper is the introduction of CrownBench, a physics-aware evaluation framework and benchmark tailored for automated dental crown design. The authors aim to shift the evaluation paradigm from clinically agnostic geometric approximation (e.g., Chamfer Distance) to clinical validity. They propose a novel set of Digital Clinical Metrics that mathematically formulate implicit clinical constraints into explicit feasibility (e.g., Occlusal/Proximal Feasibility Rates) and quality metrics. Through benchmarking, they expose a “Geometric Trap” in current state-of-the-art methods and demonstrate via a user study that their metrics align better with human expert judgments.

  • 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.

    High Clinical Relevance & Paradigm Shift: The paper addresses a critical bottleneck in the MICCAI community regarding dental restoration: the disconnect between standard CV metrics (e.g., Chamfer Distance) and actual clinical viability. Exposing the “Geometric Trap” is a highly valuable contribution that urges the community to rethink how generative models are evaluated.

    Novel Formulation of Clinical Knowledge: Translating qualitative clinical procedures, like the Virtual Articulating Paper and Digital Dental Floss, into explicit, computable mathematical formulations (OFR, PFR) is an elegant methodological innovation.

    Valuable Community Resource: The introduction of CrownBench and the promise to release a paired clinical dataset of real-world cases with expert human designs is a significant contribution that will facilitate future research.

  • 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.

    Critical Data Discrepancy and Limited Scale: There is a major contradiction regarding the dataset size that severely undermines the credibility of the experimental results. Section 3.1 states the dataset comprises 500 cases partitioned with an 8:1:1 ratio, which implies a test set of only 50 cases. However, Section 3.2 claims that Table 1 presents comparisons “across 1,000 test cases”. If the test set is indeed only 50 cases, this scale is far too small to establish a robust and reliable benchmark for the community.

    Lack of Sensitivity Analysis for Clinical Metrics: The proposed feasibility metrics heavily rely on hard-coded empirical thresholds (e.g., [-0.2mm,0.5mm] and [-0.3mm,0mm]). The authors fail to provide a sensitivity analysis. It is crucial for a benchmark to demonstrate whether slight variations in these predefined tolerance thresholds would drastically alter the ranking of the benchmarked methods.

  • 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.

  • 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

    While the paper astutely attempts to avoid the fallacy of ‘more contact area is better’ by introducing COC and OCD, these metrics may still introduce new failure modes. For instance, if the adjacent reference tooth suffers from pathological severe wear (excessive flat contact), the COC metric would penalize a healthy, point-contact crown design. Furthermore, the OCD metric is highly sensitive to the DBSCAN hyperparameters (e.g., radius); without rigorous justification, a singular, massive, and clinically unacceptable contact area could be erroneously clustered into ‘multiple valid points’, masking functional failures.

  • 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?

    I am recommending a Weak Accept for this paper. The motivation is outstanding: translating clinical constraints into computable metrics perfectly aligns with MICCAI’s focus on clinical translation. The mathematical formalization of the “Geometric Trap” is insightful. However, a benchmark paper lives and dies by its experimental rigor, which is currently lacking. The critical contradiction regarding the number of test cases (50 vs. 1000) raises significant concerns about the validity of Table 1.I am highly open to raising my score to an Accept if the authors can convincingly resolve the data discrepancy and provide the missing robustness checks in their rebuttal.

  • 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.

    After considering the manuscript and rebuttal together, I would update my score from Weak Reject to Weak Accept. The authors clarify the major reporting errors regarding the test-set size and Human Design ADF, provide missing details and inter-rater agreement for the blinded dentist study, and report threshold sensitivity analysis showing stable method rankings. These points directly address my main concerns. Some reservations remain, including the relatively small 50-case held-out test set, the serious presentation errors in the original manuscript, and the empirical nature of several thresholds/parameters. Overall, however, the paper addresses an important evaluation gap in automated crown design, and the rebuttal improves my confidence enough to support an acceptance.



Author Feedback

Contributions (R1, R3) & Advantages over similarity-based evaluation (R2). CrownBench is not intended as another crown-generation architecture, but as a clinically grounded evaluation methodology and benchmark. Defining the correct clinical target is essential: stronger models optimized toward clinically agnostic objectives may still move the field in the wrong direction. Prior works mainly report CD/HD or distance-thresholded accuracy/F1, which remain similarity to a single reference design. Occlusion/proximal analyses are usually case-level visualizations rather than scalable test-set quantification. Our study exposes this “geometric trap”: low CD can suffer from clinical failures. CrownBench translates implicit chairside criteria into explicit, interpretable, and auditable metrics. This matters because crown design is a one-to-many task: the human reference crown is not uniquely optimal and may itself require clinical adjustment. Moreover, these metrics provide the foundation for differentiable losses or reward terms for diffusion guidance or reinforcement learning.

Reporting errors in test-set size and Table 1 (R1, R2, R4). We thank reviewers for pointing out two typos: “1,000 test cases” should be “50 held-out test cases”, and Human Design ADF should be 1.00, not 100.00.We will correct both and they do not affect the experimental conclusions. Regarding scale, each case is a complete clinical molar-restoration instance with preparation jaw, opposing jaw, adjacent teeth, margin, and expert design, which is costly to collect and comparable to or larger than many prior crown-design study. The full dataset (500 cases) and evaluation code will be public as the first benchmark in automated dental restoration.

Human rating collection and agreement (R2). Our manuscript reported the blinded study, dentist experience, Likert scale, and four dimensions, but omitted collection/agreement details. We will clarify that dentists reviewed anonymized 3D visualizations including the generated crown and context, with method identity hidden and sample order randomized. For each case/dimension, the mean rating from three dentists was used as the human reference for Spearman analysis. Inter-rater agreement was high: ICC=0.868/0.874/0.884/0.844 for occlusion/proximal/margin/overall.

Metric integration and flexibility (R2). We will clarify that the overall automated score was computed by averaging the dimension-specific metrics with equal weights. For deployment, CrownBench also supports hierarchical aggregation: feasibility metrics (OFR/PFR/MSI) first serve as safety gates, since good morphology should not compensate for occlusal interference, open contact, or poor margin; quality metrics (COC/OCD/ADF) then rank feasible designs. If a single scalar is required, task-specific weights can be calibrated from expert ratings or manufacturing standards while keeping individual metrics visible. This also explains why overall correlation is lower than dimension-specific correlations: overall acceptability is multi-factorial.

Robustness, thresholds, and parameters (R2, R4). We added a compact sensitivity analysis: OFR with occlusal lower bound -0.15/-0.20/-0.25 mm and upper bound 0.45/0.50/0.55 mm, and PFR with proximal lower bound -0.25/-0.30/-0.35 mm. Absolute pass rates changed as expected, but method rankings remained unchanged, supporting that the main conclusion does not depend on one hard-coded threshold. For COC, we agree that pathological wear or unusual occlusal schemes are boundary cases where this context assumption may be less reliable. We will state this limitation and recommend condition-specific thresholds or upstream pathology screening. For DBSCAN, parameters were empirically validated on human-designed references to match clinically meaningful contact separation and avoid splitting one broad contact patch into multiple valid contacts. We will add this justification and a supplementary check.




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.

    I would recommend to address the weaknesses raised by Reviewers, especially: 1.Noticeable inconsistencies and presentation errors 2.Unclear details in the human evaluation, limited discussion of metric integration and practical robustness, and a possible error in Table 1.

  • 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.

    While reviewers before and post-rebuttal raised several key concerns, the authors addressed several concerns of two reviewers. I would recommend to include it in camera-ready.



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 reviewers showed some enthusiasm for this work being presented at the conference. Several concerns remain but the overall contributions are strong enough to justify acceptance.



back to top