pith. sign in

arxiv: 2512.08323 · v2 · submitted 2025-12-09 · 💻 cs.CV

Detecting Dental Landmarks from Intraoral 3D Scans: the 3DTeethLand challenge

Achraf Ben-Hamadou , Nour Neifar , Ahmed Rekik , Oussama Smaoui , Firas Bouzguenda , Sergi Pujades , Niels van Nistelrooij , Shankeeth Vinayahalingam
show 18 more authors
Kaibo Shi Hairong Jin Youyi Zheng Tibor Kub\'ik Old\v{r}ich Kodym Petr \v{S}illing Kate\v{r}ina Tr\'avn\'i\v{c}kov\'a Tom\'a\v{s} Moj\v{z}i\v{s} Jan Matula Jeffry Hartanto Xiaoying Zhu Kim-Ngan Nguyen Tudor Dascalu Huikai Wu and Weijie Liu Shaojie Zhuang Guangshun Wei Yuanfeng Zhou
This is my paper

Pith reviewed 2026-05-17 00:11 UTC · model grok-4.3

classification 💻 cs.CV
keywords dental landmarksintraoral 3D scanslandmark detectiondeep learningorthodontics3D point cloudschallenge benchmark
0
0 comments X

The pith

The 3DTeethLand challenge shows deep learning methods can detect 3D dental landmarks from intraoral scans with a top rank score of 0.91.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper sets up a competition called 3DTeethLand to find better ways to locate specific points on teeth using three-dimensional scans taken inside the mouth. These points matter for orthodontists who need to diagnose issues and track how treatments are working. The authors made 340 scans available to all teams and measured how well different algorithms performed at the task. The best entry reached a combined score of 0.91 along with 0.78 precision and 0.65 recall. A reader would care because reliable landmark finding could make dental care more precise and less dependent on manual measurements.

Core claim

The 3DTeethLand challenge supplies a public dataset of 340 intraoral 3D scans to benchmark 3D dental landmark detection, resulting in 49 teams competing and the winner attaining a rank score of 0.91, mean Average Precision of 0.78, and mean Average Recall of 0.65 by applying a two-stage Stratified Transformer for segmentation followed by weighted DBSCAN.

What carries the argument

A two-stage pipeline consisting of a Stratified Transformer to segment individual teeth in the 3D scan followed by weighted DBSCAN clustering to pinpoint landmark locations.

Load-bearing premise

The 340 scans include enough examples of different tooth sizes, shapes, and positions to stand in for real-world patient variety, and that the rank score plus precision and recall numbers track with improvements in actual clinical outcomes.

What would settle it

Collecting a fresh set of 200 or more intraoral scans from a new group of patients and measuring whether the winning algorithm maintains a mean average precision above 0.75 on that set.

Figures

Figures reproduced from arXiv: 2512.08323 by Achraf Ben-Hamadou, Ahmed Rekik, and Weijie Liu, Firas Bouzguenda, Guangshun Wei, Hairong Jin, Huikai Wu, Jan Matula, Jeffry Hartanto, Kaibo Shi, Kate\v{r}ina Tr\'avn\'i\v{c}kov\'a, Kim-Ngan Nguyen, Niels van Nistelrooij, Nour Neifar, Old\v{r}ich Kodym, Oussama Smaoui, Petr \v{S}illing, Sergi Pujades, Shankeeth Vinayahalingam, Shaojie Zhuang, Tibor Kub\'ik, Tom\'a\v{s} Moj\v{z}i\v{s}, Tudor Dascalu, Xiaoying Zhu, Youyi Zheng, Yuanfeng Zhou.

Figure 1
Figure 1. Figure 1: Overview of tooth structure and anatomical landmarks. (a) Illustration of the human dental [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example of the file structure of landmarks annotation. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed method by the Radboud team. The method consists of two stages [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the proposed method proposed by the YY-LAB team. The presented approach [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the proposed method proposed by the YN-LAB team: Tooth segmentation and data [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Global pipeline of the proposed approach by the IGIP-LAB team. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Overview of the proposed method by the ChohoTech team. [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Overview of the method proposed by the 3DIMLAND team. [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: mAP score for each landmark category per scan. [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: mAR score for each landmark category per scan. [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Overall visual comparison of landmark detection obtained by the six methods. Each column [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Visual comparison of hard cases for landmark detection obtained by the six methods. The first [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
read the original abstract

Teeth landmark detection is a key task in modern orthodontics, supporting advanced diagnosis, personalized treatment planning, and effective monitoring of treatment progress. However, several significant challenges may arise due to the intricate geometry of individual teeth and the substantial variations observed across different individuals. To address these complexities, the development of advanced techniques, especially through the application of deep learning, is essential for the precise and reliable detection of 3D tooth landmarks. In this context, the 3DTeethLand challenge was held in conjunction with the International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) in 2024, calling for algorithms focused on teeth landmark detection from intraoral 3D scans. This challenge introduced a publicly available dataset for 3D dental landmark detection from 340 intraoral scans, providing a standardized benchmark to evaluate state-of-the-art approaches and encouraging methodological advances toward addressing this clinically problem. A total of 49 teams participated, and 6 teams reached the final phase. The winning team achieved a rank score of 0.91, with a mean Average Precision of 0.78 and a mean Average Recall of 0.65, demonstrating a balance between precision and recall. Top teams achieved high precision with different strategies: the first-ranked team used a two-stage Stratified Transformer with segmentation and weighted DBSCAN, while the second-ranked team adopted a single-stage DGCNN with offset regression and class-specific non-maximum suppression.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript reports on the 3DTeethLand challenge organized at MICCAI 2024 for detecting dental landmarks from intraoral 3D scans. It introduces a publicly available dataset of 340 intraoral scans as a standardized benchmark, notes participation by 49 teams with 6 reaching the final phase, and presents the top results including the winning team's rank score of 0.91, mAP of 0.78, and mAR of 0.65 achieved via a two-stage Stratified Transformer with segmentation and weighted DBSCAN; the second-place entry used a single-stage DGCNN with offset regression.

Significance. If the benchmark holds, the work supplies a valuable public dataset and community benchmark for 3D dental landmark detection that can accelerate reproducible research in orthodontic applications. The reported participation numbers and diversity of top-performing strategies (transformer-based versus point-cloud CNN) provide a useful snapshot of current methodological capabilities. The explicit release of the dataset is a concrete strength that supports future method development.

major comments (3)
  1. [Introduction] Introduction: The positioning of the task as supporting 'advanced diagnosis, personalized treatment planning, and effective monitoring of treatment progress' is not backed by any reader study, downstream clinical-task evaluation, or analysis showing correlation between the reported geometric metrics (rank score, mAP, mAR) and actual clinical utility.
  2. [Dataset] Dataset section: The assertion that the 340 scans capture 'substantial variations observed across different individuals' is stated without accompanying demographic or geometric statistics (age range, ethnicity, malocclusion severity, missing teeth counts), leaving the representativeness claim unquantified.
  3. [Evaluation] Evaluation: Full details on the evaluation protocol, train/test splits, and inter-rater variability or agreement metrics for the landmark annotations are not provided, which is necessary to interpret the reliability of the headline numbers (mAP 0.78, mAR 0.65).
minor comments (2)
  1. [Abstract] Abstract: 'clinically problem' appears to be a typographical error for 'clinical problem'.
  2. [Methods] Methods descriptions of the finalist teams would benefit from explicit statements on training hyperparameters, data augmentation, and inference-time settings to improve reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments on our manuscript describing the 3DTeethLand MICCAI 2024 challenge. We address each major comment below. Revisions have been made to improve clarity and completeness where feasible within the scope of a challenge report.

read point-by-point responses

  1. Referee: [Introduction] The positioning of the task as supporting 'advanced diagnosis, personalized treatment planning, and effective monitoring of treatment progress' is not backed by any reader study, downstream clinical-task evaluation, or analysis showing correlation between the reported geometric metrics (rank score, mAP, mAR) and actual clinical utility.

    Authors: We agree that the manuscript does not contain a dedicated reader study or direct correlation analysis linking the geometric metrics to clinical endpoints. The introductory phrasing draws on established orthodontic literature regarding the role of landmark detection. In the revised version we will temper the language to present these as potential clinical applications supported by prior work and will add relevant citations. A full reader study or downstream task evaluation lies outside the scope of this benchmark paper. revision: partial

  2. Referee: [Dataset] The assertion that the 340 scans capture 'substantial variations observed across different individuals' is stated without accompanying demographic or geometric statistics (age range, ethnicity, malocclusion severity, missing teeth counts), leaving the representativeness claim unquantified.

    Authors: We acknowledge that quantitative descriptors would strengthen the dataset description. The revised manuscript will include available statistics on age range, gender distribution, malocclusion severity categories, and counts of missing teeth. Ethnicity information was not collected during acquisition for privacy reasons and will be noted as such. revision: yes

  3. Referee: [Evaluation] Full details on the evaluation protocol, train/test splits, and inter-rater variability or agreement metrics for the landmark annotations are not provided, which is necessary to interpret the reliability of the headline numbers (mAP 0.78, mAR 0.65).

    Authors: We appreciate this observation. The challenge website and supplementary material contain the full protocol and split details, but the main text will be expanded to provide a self-contained description of the evaluation protocol, exact train/test split ratios, and any computed inter-annotator agreement metrics. Where inter-rater data are limited we will state this explicitly. revision: yes

Circularity Check

0 steps flagged

Empirical benchmark report with no derivation chain

full rationale

This is a challenge summary paper that releases a dataset of 340 scans and reports external team submissions evaluated with standard detection metrics (rank score 0.91, mAP 0.78, mAR 0.65). No equations, fitted parameters, or internal derivations appear in the provided text. All performance numbers originate from independent participant algorithms rather than any self-referential construction or self-citation load-bearing step. The paper is therefore self-contained against external benchmarks and exhibits no circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper introduces no new mathematical derivations, free parameters, or postulated entities. It relies on standard computer-vision evaluation practices and existing deep-learning architectures applied to a new medical domain.

axioms (1)
  • domain assumption Standard point-cloud deep learning architectures and metrics such as mean Average Precision and mean Average Recall are appropriate for 3D dental landmark detection.
    Invoked implicitly when reporting top-team performance on the challenge.

pith-pipeline@v0.9.0 · 5720 in / 1187 out tokens · 70210 ms · 2026-05-17T00:11:27.318080+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    Angelone, A

    F. Angelone, A. M. Ponsiglione, C. Ricciardi, G. Cesarelli, M. Sansone, F. Amato, Di- agnostic Applications of Intraoral Scanners: A Systematic Review , Journal of Imaging 9 (7) (2023) 134

    work page 2023

  2. [2]

    Logozzo, E

    S. Logozzo, E. M. Zanetti, G. Franceschini, A. Kilpelä, A. Mäkynen, Recent advances in dental optics – Part I: 3D intraoral scanners for restorative dentistry, Optics and lasers in engineering 54 (2014) 203–221

    work page 2014

  3. [3]

    Eggmann, M

    F. Eggmann, M. Blatz, Recent Advances in Intraoral Scanners, Journal of Dental Re- search 103 (13) (2024) 1349–1357

    work page 2024

  4. [4]

    Woodsend, E

    B. Woodsend, E. Koufoudaki, P. A. Mossey, P. Lin, Automatic recognition of landmarks on digital dental models, Computers in Biology and Medicine 137 (2021) 104819. 5https://github.com/crns-smartvision/3DTeethLand/tree/master 27

    work page 2021

  5. [5]

    Palone, G

    M. Palone, G. A. Spedicato, L. Lombardo, Analysis of tooth anatomy in adults with idealocclusion: Apreliminarystudy, AmericanJournalofOrthodonticsandDentofacial Orthopedics 157 (2) (2020) 218–227

    work page 2020

  6. [6]

    Bolelli, L

    F. Bolelli, L. Lumetti, S. Vinayahalingam, M. Di Bartolomeo, A. Pellacani, K. March- esini, N. Van Nistelrooij, P. Van Lierop, T. Xi, Y. Liu, et al., Segmenting the Inferior Alveolar Canal in CBCTs Volumes: The ToothFairy Challenge, IEEE Transactions on Medical Imaging (2024)

    work page 2024

  7. [7]

    Y. Wang, Y. Zhang, X. Chen, S. Wang, D. Qian, F. Ye, F. Xu, H. Zhang, R. Dan, Q.Zhang, etal., MICCAI2023STSChallenge: Aretrospectivestudyofsemi-supervised approaches for teeth segmentation, Pattern Recognition (2025) 112049

    work page 2025

  8. [8]

    Ben-Hamadou, O

    A. Ben-Hamadou, O. Smaoui, A. Rekik, S. Pujades, E. Boyer, H. Lim, M. Kim, M. Lee, M. Chung, Y.-G. Shin, et al., 3DTeethSeg’22: 3D Teeth Scan Segmentation and Label- ing Challenge, arXiv preprint arXiv:2305.18277 (2023)

    work page arXiv 2023

  9. [9]

    A.Rekik, A.Ben-Hamadou, O.Smaoui, F.Bouzguenda, S.Pujades, E.Boyer, TSegLab: Multi-stage 3D dental scan segmentation and labeling, Computers in Biology and Medicine 185 (2025) 109535

    work page 2025

  10. [10]

    Triarjo, R

    S. Triarjo, R. Sarno, S. C. Hidayati, G. Sihaj, Automatic 3D Digital Dental Landmark Based on Point Transformation Weight , in: 2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC), IEEE, 2023, pp. 336–341

    work page 2023

  11. [11]

    T.-H. Wu, C. Lian, S. Lee, M. Pastewait, C. Piers, J. Liu, F. Wang, L. Wang, C.-Y. Chiu, W. Wang, et al., Two-Stage Mesh Deep Learning for Automated Tooth Seg- mentation and Landmark Localization on 3D Intraoral Scans , IEEE transactions on medical imaging 41 (11) (2022) 3158–3166

    work page 2022

  12. [12]

    G. Wei, Z. Cui, J. Zhu, L. Yang, Y. Zhou, P. Singh, M. Gu, W. Wang, Dense Rep- resentative Tooth Landmark/axis Detection Network on 3D Model , Computer Aided Geometric Design 94 (2022) 102077

    work page 2022

  13. [13]

    Y. Lang, X. Chen, H. H. Deng, T. Kuang, J. C. Barber, J. Gateno, P.-T. Yap, J. J. Xia, DentalPointNet: landmark localization on high-resolution 3D digital dental models, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2022, pp. 444–452

    work page 2022

  14. [14]

    Y. Lang, H. H. Deng, D. Xiao, C. Lian, T. Kuang, J. Gateno, P.-T. Yap, J. J. Xia, DLLNet: An Attention-Based Deep Learning Method for Dental Landmark Localiza- tion on High-Resolution 3D Digital Dental Models, in: Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Stras- bourg, France, September 27–Octob...

    work page 2021

  15. [15]

    Kubík, M

    T. Kubík, M. Španěl, Robust Teeth Detection in 3D Dental Scans by Automated Multi- View Landmarking, in: Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies - BIOIMAGING, SciTePress, 2022, pp. 24–34

    work page 2022

  16. [16]

    C. Lian, L. Wang, T.-H. Wu, F. Wang, P.-T. Yap, C.-C. Ko, D. Shen, Deep Multi- Scale Mesh Feature Learning for Automated Labeling of Raw Dental Surfaces From 3D Intraoral Scanners , IEEE transactions on medical imaging 39 (7) (2020) 2440–2450

    work page 2020

  17. [17]

    C. Lian, L. Wang, T.-H. Wu, M. Liu, F. Durán, C.-C. Ko, D. Shen, MeshSNet: Deep Multi-scale Mesh Feature Learning for End-to-End Tooth Labeling on 3D Dental Sur- faces, in: Medical Image Computing and Computer Assisted Intervention – MICCAI 2019, Springer, 2019, pp. 837–845

    work page 2019

  18. [18]

    Teeth3DS: A benchmark for teeth segmentation and labeling from intra-oral 3D scans,

    A.Ben-Hamadou, N.Neifar, A.Rekik, O.Smaoui, F.Bouzguenda, S.Pujades, E.Boyer, E. Ladroit, Teeth3DS+: An Extended Benchmark for Intraoral 3D Scans Analysis , arXiv preprint arXiv:2210.06094 (2022)

    work page arXiv 2022

  19. [19]

    Ester, H.-P

    M. Ester, H.-P. Kriegel, J. Sander, X. Xu, A density-based algorithm for discovering clusters in large spatial databases with noise, in: Proceedings of the Second Interna- tional Conference on Knowledge Discovery and Data Mining, AAAI Press, 1996, p. 226–231

    work page 1996

  20. [20]

    Garland, P

    M. Garland, P. S. Heckbert, Surface Simplification Using Quadric Error Metrics, in: Proceedings of the 24th annual conference on Computer graphics and interactive tech- niques, ACM Press/Addison-Wesley Publishing Co., 1997, pp. 209–216

    work page 1997

  21. [21]

    Zheng, B

    Y. Zheng, B. Chen, Y. Shen, K. Shen, TeethGNN: Semantic 3D Teeth Segmenta- tion With Graph Neural Networks, IEEE Transactions on Visualization and Computer Graphics 29 (7) (2022) 3158–3168

    work page 2022

  22. [22]

    H. Jin, Y. Shen, J. Lou, K. Zhou, Y. Zheng, KeypointDETR: An End-to-End 3D Keypoint Detector, in: European Conference on Computer Vision, Springer, 2025, pp. 374–390

    work page 2025

  23. [23]

    Bengio, J

    Y. Bengio, J. Louradour, R. Collobert, J. Weston, Curriculum learning, in: Proceedings of the 26th annual international conference on machine learning, 2009, pp. 41–48

    work page 2009

  24. [24]

    Dascalu, B

    T. Dascalu, B. Ibragimov, Assignment Theory-Augmented Neural Network for Den- tal Arch Labeling, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2023, pp. 295–304

    work page 2023

  25. [25]

    J. Li, J. Zhou, Y. Xiong, X. Chen, C. Chakrabarti, An Adjustable Farthest Point Sam- pling Method for Approximately-sorted Point Cloud Data, in: 2022 IEEE Workshop on Signal Processing Systems (SiPS), IEEE, 2022, pp. 1–6. 29

    work page 2022

  26. [26]

    X. Ma, C. Qin, H. You, H. Ran, Y. Fu, Rethinking network design and local geometry in point cloud: A simple residual MLP framework, arXiv preprint arXiv:2202.07123 (2022)

    work page arXiv 2022

  27. [27]

    Tran, T.-D

    T.-H. Tran, T.-D. Cao, T.-T.-H. Tran, HDBSCAN: Evaluating the Performance of Hierarchical Clustering for Big Data, in: Soft Computing: Biomedical and Related Applications, Springer, 2021, pp. 273–283

    work page 2021

  28. [28]

    Y. You, et al., KeypointNet: A Large-scale 3D Keypoint Dataset Aggregated from Numerous Human Annotations , in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2020, pp. 13644–13653

    work page 2020

  29. [29]

    Maier-Hein, A

    L. Maier-Hein, A. Reinke, P. Godau, M. D. Tizabi, F. Buettner, E. Christodoulou, B. Glocker, F. Isensee, J. Kleesiek, M. Kozubek, et al., Metrics reloaded: Recommen- dations for image analysis validation , Nature methods 21 (2) (2024) 195–212. 30

    work page 2024