List of Papers Browse by Subject Areas Author List
Abstract
Dental implantation restores missing teeth through surgical insertion of artificial roots, relying on preoperative digital planning to ensure precision and efficiency. However, critical challenges persist in virtual tooth positioning: this process demands extensive clinical expertise and time-consuming manual adjustments due to ambiguous anatomical references from missing teeth. To address these limitations, we propose a unified framework that accurately predicts three dimensional (3D) shapes and positions of missing teeth in diverse patterns, enabling anatomy-aware preoperative planning. Our proposal introduces two technical innovations: (1) A dynamic iterative generation strategy is proposed to progressively predict multiple missing teeth one by one using a target tooth identification module, accommodating arbitrary tooth loss patterns without case-specific retraining; (2) A tooth-centroid-prompted conditional diffusion model is developed to leverage geometric constraints from predicted tooth centroid and adjacent teeth to generate high-fidelity point cloud reconstructions. Extensive experiments show that our model outperforms conventional deep learning baseline in predicting multiple missing teeth, achieving a prediction accuracy of 1.30mm (Chamfer Distance) and an angular error of 5.42 degrees. This improvement has the potential to enhance the accuracy and efficiency of dental implant planning by providing precise anatomical references for missing teeth, potentially revolutionizing digital dentistry workflows.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2258_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{JiZon_3D_MICCAI2025,
author = { Ji, Zongrui AND Li, Na AND Xue, Peng AND Dong, Yi AND Ma, Lei},
title = { { 3D Dynamic Prediction of Missing Teeth in Diverse Patterns via Centroid-prompted Diffusion Model } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15968},
month = {September},
page = {3 -- 12}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper under review proposes a novel approach for predicting missing teeth. The frameworks support the prediction of multiple missing teeth and add a conditional diffusion model to take into account geometric constraints when proposing a point cloud for a tooth. The method is then extensively tested in the context of preoperative planning that is flexible enough and can adapt to a particular tooth geometry for a given patient.
- 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.
-very clear clinical application and well validated approach with several cases presented. -end to end framework that can generate patient adapted tooth geometries and takes into account the neighboring teeth when proposing an alternative for the implant. -very well written paper
- 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.
-somewhat limited technical novelty since the diffusion methods are adopted from a previously published paper, however they are adapted to the task at hand -more ablation studies related to the number of points required in the point cloud that are needed for acceptable results are needed.
- 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.
- 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/2025/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 — could be 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 presents a clinically useful CAD tool that can improve the planning process of tooth implants. The dataset include 130 healthy patients which is then used to simulate different tooth loss patterns. Both single and multiple tooth loss patterns are investigated. Results are compared against ground truth and the proposed method compared against a simple U-net. The method shows potential for data driven personalized implantology.
- 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 authors have clarified some issues raised and even though the deep learning methods are not novel the framework they propose is and it is a valuable contribution as an application paper.
Review #2
- Please describe the contribution of the paper
Tooth positioning during dental implantation is a challenging task. The paper proposes a framework for accurately predicting 3D shapes and positions of missing teeth. It consists of a two-stage process repeated until all missing teeth have been identified. Stage 1 identifies the centroid of a missing tooth which has a high number of adjacent teeth either present or planned. Stage 2 uses point-cloud diffusion to predict the 3D shape of the missing tooth. The model is evaluated extensively on a dataset of 3D dental scans, achieving a Chamfer Distance (CD) of 1.30mm and an angular error 5.42 degrees.
- 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 manuscript is clearly-motivated and well-written, describing difficulties inherent in the domain and how the proposed method solves them.
- The proposed approach is a novel, comprehensive solution for the problem of missing teeth prediction.
- The use of a point-cloud diffusion model is justified by comparison with a volumetric U-Net.
- 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.
- Straw-man baseline. A volumetric U-Net is inherently limited by the resolution of the voxel grid.
- Additional clarity is needed regarding the baseline. Is it also a diffusion model?
- 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.
- 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/2025/en/REVIEWER-GUIDELINES.html
The comparison to the volumetric U-Net is weak. The U-Net used a voxel grid of 144x144x224, but it is not stated what the spatial resolution of this grid is. This approach has a minimum accuracy that corresponds to the voxel size and may therefore not be a fair comparison. This is evident in the visualizations of the U-Net predictions, which are substantially smoothed (with what parameters are not stated). Several of the U-Net predictions appear to have large flat areas, resulting from the boundary of the voxel grid.
The authors should provide more details for the U-Net baseline. Is it a diffusion model? What is the spatial resolution? How much of the surrounding teeth structure is it conditioned on, if so? What parameters were used for smoothing? This is especially important if the meshes are used to compute the CD and Hausdorff distances, as the parameters would affect the results.
Minor comments:
- The manuscript may benefit from more readable table design, e.g. as outlined in https://people.inf.ethz.ch/markusp/teaching/guides/guide-tables.pdf.
- 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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although the manuscript compares to a straw man baseline, the proposed method is a novel and comprehensive solution to the problem of missing teeth prediction. The authors have provided a clear motivation for the work, and the results are sufficient to warrant publication. However, the authors should clarify the baseline and provide more details on the U-Net model and post-processing parameters, such as smoothing.
- 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.
The authors have addressed my concerns, and I believe additional experiments are not warranted (and would be prohibited), but I am unsure how to evaluate the points raised by R3’s expertise on artificial roots and implants. But I believe the rebuttal makes a solid case for reconstructing the original tooth shape, which should be integrated into the final version.
The U-Net baseline resolution is acceptable with 0.4 mm isotropic spacing compared to average errors well above 1 mm, so that is not the botteleneck.
The desire for an ablation study on the number of points is reasonable but not necessary for publication, as it is clear that the current number of points is sufficient to achieve the desired results.
Review #3
- Please describe the contribution of the paper
The main contribution of the paper is the synthetic generation of anatomically plausible artificial teeth at missing positions using a diffusion-based model.
- 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.
- Enables clinically meaningful crown generation in edentulous regions to support preoperative planning
- Leverages and adapts state-of-the-art diffusion models for the synthesis of realistic and context-aware dental anatomy
- 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 method does not model or reconstruct artificial roots, as current clinical practice relies on implants rather than full artificial teeth.
- The spatial relationship between implant and crown is not modeled, limiting applicability to root-level planning.
- Jawbone quality, a critical factor for implant success, is not incorporated into the prediction or generation process.
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Limited clinical applicability, as the method focuses on technical image-based generation without integrating real-world clinical constraints
- 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
Author Feedback
- Response to the concern of technical novelty of this study (R1) We would like to clarify that while our framework adopts the diffusion model architecture from prior work, our primary technical contributions reside in the development of a novel application framework for predicting the original morphology of missing teeth including: (1) A dynamic iterative generation strategy is proposed to progressively predict multiple missing teeth one by one using a target tooth identification module, accommodating arbitrary tooth loss patterns without case-specific retraining. (2) A tooth-centroid prompting mechanism is developed for multiple missing teeth prediction to leverage geometric constraints from predicted tooth centroid.
- Response to the ablation study related to the point number(R1) We totally agree that a detailed ablation study on point cloud resolution is essential for understanding the trade-offs between computational efficiency and reconstruction accuracy. However, due to space limitations in MICCAI manuscript, we had to focus on presenting the core technical contributions and validation results of this study. In the future, we will perform comprehensive ablation studies and include them in the final version of this work.
- Response to concerns related to the baseline U-Net (R2) The U-Net baseline is a traditional volumetric convolutional network, not a diffusion model. It operates on voxelized 3D image and directly predicts missing tooth volumes. Our choice of U-Net as a baseline was motivated by its widespread adoption in medical image processing and 3D shape reconstruction tasks ([12–14] in the manuscript). While we acknowledge that voxel-based methods can be constrained by resolution, we clarify that the U-Net baseline in our experiments utilized a high-resolution volumetric grid of 0.4×0.4×0.4mm, which already minimizes discretization errors to a clinically acceptable level. However, the primary limitation of U-Net lies not in resolution but in its inherent inability to learn complex geometric features, even with high-resolution inputs. This geometric learning deficiency is explicitly addressed by our proposed point cloud diffusion framework, which directly models 3D surfaces as continuous point distributions.
- Response to concerns related to the evaluation results (R2) 1). Smoothing in U-Net Predictions The “smoothing” observed in U-Net predictions arises from marching cube (smooth iterations=15) based mesh reconstruction in visual comparisons. To avoid biases introduced by mesh generation, the marching cube parameters were identical for U-Net and our method (in our point cloud acquisition process). 2). The large flat areas observed in U-Net’s predictions The large flat areas observed in U-Net predictions (Figs. 2–3) primarily stem from the limited geometric learning capability of the U-Net framework: (1) As shown in Fig.2, U-Net’s voxel-based architecture struggled to model thin or intricate structures, i.e., the tooth roots, due to its reliance on discrete grid representations. (2) In Fig.3, U-Net failed to reconstruct fine anatomical details within the predefined spatial bounds of the input image, which leads to the large flat areas observed in U-Net’s predictions.
- Response to absence of artificial root modeling and jawbone quality factor (R3) Our study aims to reconstruct the original 3D geometry of missing natural teeth (i.e., their pre-loss morphology), rather than directly predicting artificial roots or implants. The predicted pre-loss position and morphology serve as foundational references for implant placement. We didn’t consider jawbone quality because it introduces variability due to post-loss bone remodeling (e.g., alveolar resorption). Bone resorption alters the spatial relationship between residual bone and the original roots, complicating direct correlation with pre-loss tooth morphology. Integrating this factor into our framework may lead to uncertainty.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
Reviewers agree that this is a valuable contribution with an impact for clinical translation.
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A