List of Papers Browse by Subject Areas Author List
Abstract
Tooth alignment is a fundamental procedure in orthodontic treatment, requiring prolonged stage-wise clinical evaluation to precisely manage complex tooth movements. Conventional automatic methods regress a direct rigid transformation in a one-shot manner and impose clinical constraints only on the final configuration, limiting the ability to reflect the progressive nature of actual orthodontic treatment and yielding unreliable outcomes in complex tooth movement cases. To address this limitation, we propose a framework that models a virtual progressive tooth movement trajectory in which clinical constraints can be evaluated at each stage. The framework decomposes alignment into K incremental rigid-transformation steps and generates a stable and continuous virtual tooth trajectory from pre-/post-treatment data alone. By enforcing clinical constraints at every intermediate stage, the framework models progressive tooth movements in orthodontic treatment, demonstrating more precise and reliable tooth alignment outcomes. Our code is publicly available at https://github.com/moontaijin/ProMoT-Tooth-Alignment.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5316_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/moontaijin/ProMoT-Tooth-Alignment
Link to the Dataset(s)
TeethAlign3D Dataset v1 : https://zenodo.org/records/15834700
BibTex
@InProceedings{JoSeu_Tooth_MICCAI2026,
author = { Jo, SeungKwan AND Moon, Taejin AND Kim, So Hyun AND Jang, Hyensoo AND Chung, Minyoung},
title = { { Tooth Alignment via Virtual Trajectories with Clinical Constraints } },
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
The paper proposes a framework for tooth alignment that models alignment as a sequence of incremental rigid transformations rather than a one-shot prediction. By decomposing the task into a multi-step trajectory and enforcing clinically motivated constraints at intermediate stages, the method introduces a structured way to incorporate domain knowledge. Additional components, including trajectory stabilization and intermediate-state self-learning, are designed to improve the predicted transformations.
- 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 paper presents a well-designed approach to tooth alignment, with several strengths:
First, the formulation of tooth alignment as a multi-step trajectory prediction problem is an intuitive departure from standard one-shot regression approaches. By decomposing large transformations into incremental steps, the method reflects the underlying structure of the problem and leads to improved performance.
Second, the work stands out in its use of domain-informed modeling. In particular, the incorporation of clinically motivated constraints at intermediate stages is a well-motivated design choice that aligns the optimization process with the clinical properties of tooth movement. This is further strengthened by the training strategy, including trajectory stabilization and intermediate-state self-learning. Together, these components form a coherent framework where both the objective and training are tailored to the problem.
Finally, the method shows consistent quantitative improvements over compared baselines across multiple metrics. Overall, the paper combines a solid methodological design, and a convincing empirical performance.
- 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.
First, some of the claims are overstated. In particular, the paper claims to “mimic the stage-wise decision-making process” of clinical orthodontic treatment. However, the proposed method does not model decision-making in a meaningful sense: it does not include any mechanism for action selection, adapting decisions based on intermediate outcomes, or explicitly representing clinical reasoning. Instead, it enforces predefined constraints during optimization. While this introduces a process-like structure, it should not be equated with mimicking clinical decision-making, and the claim should be toned down.
Second, several component-level interpretations are insufficiently supported by the presented experiments. For example, the paper attributes roles to individual components, such as TSS “stabilizing early trajectory learning” or self-learning “densifying trajectories. ” However, the experimental results (e. g. , ablation study) mainly demonstrate improvements in final performance metrics, without directly validating these more detailed behavioral claims. As a result, some of these statements read as the authors’ interpretations rather than empirically established effects.
Third, some design choices are introduced without sufficient motivation. For instance, the number of trajectory steps (K=21) is a central parameter but is not justified. Similarly, the decision to apply trajectory stabilization only during early training is not explained. Providing intuition or empirical justification for these aspects would improve clarity.
- 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
Some minor points that could improve presentation:
*The keyword “Scan” is overly generic; consider using a more specific term (e. g. , “Intraoral Scan (IOS)” or “3D Dental Scans”). *Fig. 1 is visually clean but not very informative; as an early overview, it would benefit from a clearer depiction of the data flow and intermediate stages while maintaining simplicity. *There are some inconsistencies in notation, e. g. , reuse of E (encoder vs. tooth-pair set), and mismatches between Fig. 2 and its caption (e. g. , x/X, indexing/notation of transformation T). Unifying notation would improve readability. *Some sentences could be improved for clarity and grammar, e. g. : **“enabling stage-wise evaluation direct pre/post dataset” appears to be missing a preposition, **repetition of “we introduce” in the contributions list, **ambiguity in “that yields stable and continuous…” (unclear what it refers to). *Some equations could benefit from improved LaTeX formatting (e. g. , properly scaled parentheses). *Minor punctuation issue: Eq. (6) ends with a period although the sentence continues.
- 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?
The paper presents a well-designed and well-motivated approach to tooth alignment. Modeling alignment as a progressive trajectory and incorporating clinically motivated constraints at intermediate stages provides an intuitive alternative to standard one-shot regression approaches. The training strategy is carefully constructed to reflect domain-specific aspects of tooth movement. The method is supported by consistent improvements over compared baselines and an ablation study.
At the same time, some aspects of the paper would benefit from more careful positioning. In particular, the claim of “mimicking stage-wise clinical decision-making” is overstated. Similarly, some interpretations of component behavior are not directly validated and should be presented more cautiously. Overall, I find the core idea and execution strong and believe the paper meets the bar for acceptance. The main issues are related to overclaiming rather than fundamental flaws in the method, and can be addressed with relatively minor revisions.
- Reviewer confidence
Very confident (4)
- [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 #2
- Please describe the contribution of the paper
This paper studies automatic tooth alignment from intraoral scan data and formulates the task as a staged prediction problem rather than a one-shot rigid transformation regression. Instead of directly predicting the final transform for each tooth, the method predicts a virtual K-step trajectory of incremental SE(3) updates using an SSM-based sequential regressor. Clinical constraints, including collision avoidance, occlusal plane consistency, and spacing control, are enforced at each stage.
To enable training using only pre- and post-treatment pairs, the paper introduces a trajectory stabilization strategy based on uniform movement regularization and geodesic pseudo-targets, together with an intermediate-state self-learning scheme.
- 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 presents a process-centric framework for automatic tooth alignment, replacing one-shot endpoint regression with staged SE(3) refinement under intermediate clinical constraints. The idea is interesting and well motivated, and the proposed approach achieves competitive performance compared with the reported baselines. The ablation study is also informative and suggests that a substantial portion of the performance improvement arises from the progressive modeling strategy.
- 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.
My main concern is that the learned “trajectory” remains a virtual trajectory derived from pre-/post-treatment pairs, rather than a clinically validated treatment trajectory. The intermediate supervision is constructed through geodesic pseudo-targets, stage-wise constraints, and self-learning signals. As a result, while the paper demonstrates improved staged refinement behavior, it does not yet clearly establish that the model captures real clinical treatment dynamics. The corresponding claims should therefore be moderated to better reflect this limitation.
A second concern relates to the experimental comparison. The current evaluation is still largely dominated by one-shot regression baselines. While this setup supports the argument that progressive modeling can outperform one-shot prediction, it does not fully demonstrate competitiveness with more recent trajectory-oriented or diffusion-based approaches. In addition, the proposed trajectory stabilization strategy relies on several strong priors, such as relatively uniform per-stage movements and geodesic pseudo-targets, whose clinical realism and practical justification are not yet fully discussed.
- 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
Nan
- 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?
Nan
- 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
Review #3
- Please describe the contribution of the paper
1.This work redefines the traditional one-shot tooth alignment problem as a process-centric, K-step incremental 6-DoF trajectory prediction task. Unlike existing methods that directly regress a final transformation, the proposed formulation enables the model to learn progressive movement trajectories using only pre-treatment and post-treatment paired data. This reformulation further allows clinical evaluation to be performed at every intermediate stage, which is essential for mimicking real orthodontic workflows. 2.The paper introduces several methodological innovations to enable stable and continuous virtual trajectory learning. A state-space model is adopted as the sequential predictor to capture long-range dependencies over the different refinement steps. A trajectory stabilization strategy is proposed, which includes per-stage movement distribution regularization and geodesic pseudo-target supervision, effectively preventing trajectory collapse. In addition, a progressive self-learning mechanism is designed to reuse and retrain on intermediate states, producing denser and smoother movement dynamics. 3.Unlike prior methods that impose clinical constraints only on the final alignment, the proposed framework enforces constraints such as collision avoidance, occlusal plane preservation, and interproximal spacing control at every intermediate stage. This stage-wise enforcement directly mimics the clinician’s decision-making process of iterative evaluation, fine adjustment, and accumulation. As a result, the framework achieves more reliable and precise tooth alignment, especially in complex cases involving irregular multi-tooth movements.
- 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.
1.Novel Progressive Trajectory Modeling Paradigm. Existing methods mostly regress the final transformation directly, ignoring the clinical importance of intermediate stages. This paper is the first to formulate tooth alignment as a multi-step SE(3) trajectory and achieve progressive movement through cumulative composition. This paradigm is pioneering in automated orthodontic planning. 2.Clever Design of Trajectory Stabilization Strategy. Training a multi-step trajectory using only pre-treatment and post-treatment data can easily lead to trajectory collapse, such as excessively large single-step displacements or failure to converge. The proposed TSS addresses this problem through uniform movement regularization, which encourages all teeth to move equally at each step, and geodesic pseudo-target interpolation, which provides soft supervision for intermediate steps. Moreover, TSS is applied only in the early training phase using a cosine schedule, so it does not interfere with later fine learning. 3.Use of Only Pre-treatment and Post-treatment Data. Many clinical datasets contain only the initial and final scans without intermediate stage annotations. The proposed method does not rely on intermediate labels, which gives it high practicality and strong generalization potential.
- 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: Lack of validation against real clinical intermediate states. The paper does not use any real patient intermediate scans to verify whether the generated virtual trajectories match actual tooth movement. It is recommended to include even a small amount of real intermediate data for qualitative comparison or cite existing biomechanical studies. 2.The choice of K=21 lacks justification. The paper fixes K=21 without providing clinical or experimental grounds for this value. .
- 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.
(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?
This paper presents a novel process-centric framework for tooth alignment by modeling virtual progressive trajectories with stage-wise clinical constraints. The use of state space models, trajectory stabilization, and intermediate self-learning is technically sound and shows clear performance improvements over existing methods on a public dataset. The idea of enforcing clinical constraints at every intermediate stage is clinically relevant and distinguishes this work from one-shot regression approaches. However, the paper has notable limitations, including the lack of validation against real intermediate scans and the absence of justification for the choice of K=21.These weaknesses reduce confidence in the clinical plausibility of the generated trajectories and the generalizability of the method. Nevertheless, the core contributions are valuable and the experimental results are promising.
- 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 the reviewers and meta-reviewer for their constructive comments. We appreciate that the clinical relevance of process-centric tooth alignment and progressive virtual trajectories with stage-wise constraints were recognized.
[On clinical decision-making — R1, R2] We agree that “mimicking stage-wise clinical decision-making” should be interpreted carefully. Our method does not model full clinical reasoning, action selection, or clinician-specific policies. Rather, it approximates a process-centric alignment workflow by evaluating collision avoidance, occlusal-plane consistency, and spacing control at each virtual stage, not only at the final configuration. We will moderate this wording with expressions such as “process-centric alignment with stage-wise evaluation” and “virtual progressive tooth movement trajectory with intermediate-stage clinical constraints,” and clarify that such claims indicate a long-term direction, not a fully realized capability.
[On virtual trajectories — R2, R3] We acknowledge that generated trajectories are learned from pre-/post-treatment pairs, not clinically observed intermediate trajectories. As noted in the Introduction, most public/real-world datasets provide only pre-/post-treatment pairs; thus, our goal is a practical framework under this data constraint, not recovery of true patient-specific biological movement. It decomposes large one-shot transformations into progressive SE(3) updates and enables intermediate constraint enforcement from only pre-/post-treatment data. We will clarify that refined trajectory modeling with intermediate scans/data remains future work.
[On dataset/failure cases — Meta-review, R2, R3] We acknowledge that evaluation on diverse malocclusion types, jaw morphologies, and difficult cases would strengthen generalizability. Our evaluation uses the public paired pre-/post-treatment IOS dataset with 5-fold cross-validation and geometric/motion metrics, but intermediate scans and extreme scenarios remain beyond scope. We will state these limitations and discuss the need for further analyses, including severe crowding and anchorage-critical cases.
[On K=21 and early TSS — R1, R3] We will better justify K=21.It was selected as a staged refinement length motivated by prior clear-aligner studies reporting sequences around this scale, while keeping rollout tractable. It is not an exact clinical treatment schedule. TSS is mainly applied in early training because geodesic pseudo-targets and movement regularization provide an initial scaffold; later training is driven by final alignment and stage-wise constraints. We will clarify this and cite prior aligner studies.
[On component interpretation — R1] Component-level roles should be stated cautiously. The ablation study shows final-performance gains from progressive modeling, stage-wise constraints, TSS, and self-learning, but does not directly prove all behavioral interpretations. We will revise statements such as “TSS stabilizes early trajectory learning” and “self-learning densifies trajectories” to clarify that these are design motivations and interpretations supported by ablation trends and visualization, not established clinical effects.
[On comparison scope — R2] Comparisons with recent diffusion-based methods would be valuable references. However, our main contribution is not a stronger architecture, but a paradigm shift from one-shot endpoint prediction to progressive trajectory prediction with stage-wise constraints. We will clarify this scope and discuss model-level extensions without overstating competitiveness.
[On presentation/reproducibility — R1, R2] We will improve clarity by unifying notation, correcting grammar/LaTeX issues, clarifying Figs. 1/2, replacing “Scan,” and adding implementation details. We will also release code to support reproducibility.
We appreciate the reviewers’ feedback and will improve clarity, positioning, and clinical interpretation.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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.
This paper presents a novel formulation for tooth alignment that plans virtual trajectories under clinically meaningful constraints, most notably the avoidance of inter-tooth collisions during the simulated alignment process. The key contribution is the introduction of a trajectory-based optimization framework that explicitly models the biomechanical and spatial constraints inherent in orthodontic treatment planning, moving beyond static target pose estimation. Reviewer R1 characterized this work as “a standout contribution” with a “clear, well-executed idea that produces strong results on a challenging problem,” while R2 acknowledged the “clinical relevance and the clear improvement over existing methods.”
Both reviewers recognized the technical soundness and practical importance of the approach. R1’s score of 5 (Accept) reflects enthusiasm for the novelty and execution, noting that the method addresses a real gap in current computational orthodontics by ensuring that planned tooth movements are physically plausible and collision-free throughout the entire treatment sequence. R2, assigning a score of 4 (Weak Accept), similarly appreciated the clinical framing and the demonstrable performance gains over baseline approaches, which supports the paper’s value to the MICCAI community.
The primary limitation, as noted by R2, concerns the evaluation dataset, which was described as “somewhat limited.” While the results are convincing within the presented scope, a larger and more diverse set of clinical cases (e.g., varying degrees of malocclusion, different jaw morphologies) would strengthen the generalizability claims. Additionally, neither reviewer deeply interrogated potential failure cases, such as extreme crowding or anchorage-critical scenarios, which remain open questions for future work. Nonetheless, these limitations do not undermine the core contribution, which is well-supported by the evidence provided.
Given the strong average score of 4.5, the clear enthusiasm from R1, and the overall positive assessment from R2 regarding clinical relevance and methodological improvement, there is a clear consensus that this work meets the high bar for acceptance at MICCAI 2026.The novelty of formulating tooth alignment as a trajectory planning problem under clinical constraints is compelling, and the execution demonstrated is rigorous enough to warrant publication. I therefore recommend unconditional acceptance.
