pith. sign in

arxiv: 2507.22512 · v2 · submitted 2025-07-30 · 💻 cs.CV · cs.LG· eess.IV

AlphaDent: A dataset for automated tooth pathology detection

Evgeniy I. Sosnin , Yuriy L. Vasilev , Roman A. Solovyev , Aleksandr L. Stempkovskiy , Dmitry V. Telpukhov , Artem A. Vasilev , Aleksandr A. Amerikanov , Aleksandr Y. Romanov This is my paper

Pith reviewed 2026-05-19 02:59 UTC · model grok-4.3

classification 💻 cs.CV cs.LGeess.IV
keywords dental datasettooth pathologyinstance segmentationneural networkdental imagingautomated detectionopen datasetDSLR photographs
0
0 comments X

The pith

A new dataset of over 1200 tooth photographs from 295 patients enables high-quality neural network predictions for instance segmentation of dental pathologies.

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

The paper introduces AlphaDent, a dataset built from DSLR camera photographs of teeth belonging to 295 patients that totals more than 1200 images. These images carry labels for the instance segmentation task across nine classes of tooth conditions. The authors describe the dataset construction, the labeling format used, and the results of an experiment that trains a neural network on the data for this segmentation problem. The experiment shows that the trained network produces high quality predictions. The full dataset, training code, and model weights are released under open licenses.

Core claim

The AlphaDent dataset consists of over 1200 DSLR photographs of teeth from 295 patients that have been labeled into nine classes for instance segmentation, and neural networks trained on it deliver high quality predictions for automated tooth pathology detection.

What carries the argument

Instance segmentation labels across nine tooth pathology classes on the collected DSLR images, which provide the training signal for neural networks to locate and outline individual pathologies.

If this is right

  • Researchers can train and evaluate instance segmentation models that achieve high quality results on dental photographs.
  • The open dataset and released code allow direct reproduction of the reported prediction quality.
  • Automated tools built from this data can identify tooth conditions directly from standard camera images.
  • The nine-class labeling scheme supplies a concrete starting point for further model development in dental analysis.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The dataset could support training of models that assist initial screening from everyday photos before clinical visits.
  • Combining the labels with other dental image types might improve overall detection across different capture conditions.
  • Community extensions of the nine classes could address additional pathologies not covered in the initial release.

Load-bearing premise

The manual labeling of images into the nine classes is accurate, consistent, and representative of real-world tooth pathologies.

What would settle it

A neural network trained on the dataset produces low-accuracy segmentation masks that fail to match independent expert labels on a new collection of tooth photographs.

Figures

Figures reproduced from arXiv: 2507.22512 by Aleksandr A. Amerikanov, Aleksandr L. Stempkovskiy, Aleksandr Y. Romanov, Artem A. Vasilev, Dmitry V. Telpukhov, Evgeniy I. Sosnin, Roman A. Solovyev, Yuriy L. Vasilev.

Figure 1
Figure 1. Figure 1: Examples of photographs of the oral cavity with markings [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: shows the examples of markup for each class [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of patients by age [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Yolo v8 neural network architecture Training settings: 9 classes. Task type: Instance Segmentation. Image sizes are 640px and 960px. Number of epochs: 100. Yolo version: v8x Large set of augmentations. Target metric: average mAP50 across all 9 classes [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Examples of neural network predictions on validation images The code we used for training, as well as the code for inference and pre-trained weights, are available here [24]: https://github.com/ZFTurbo/AlphaDent. 7. CONCLUSION Thus, we present a new dental image dataset (AlphaDent). This is a unique dataset based on the DSLR dental photographs and contains over 1200 images for 295 patients. The dataset is … view at source ↗
read the original abstract

In this article, we present a new unique dataset for dental research - AlphaDent. This dataset is based on the DSLR camera photographs of the teeth of 295 patients and contains over 1200 images. The dataset is labeled for solving the instance segmentation problem and is divided into 9 classes. The article provides a detailed description of the dataset and the labeling format. The article also provides the details of the experiment on neural network training for the Instance Segmentation problem using this dataset. The results obtained show high quality of predictions. The dataset is published under an open license; and the training/inference code and model weights are also available under open licenses.

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

2 major / 1 minor

Summary. The manuscript introduces AlphaDent, a new dataset of over 1200 DSLR photographs of teeth from 295 patients, annotated for instance segmentation across 9 pathology classes. It details the data collection and labeling format, reports baseline neural network training experiments for instance segmentation that yield high-quality predictions, and releases the dataset, training code, and model weights under open licenses.

Significance. If the ground-truth annotations prove reliable, AlphaDent would provide a useful open resource for computer vision research on dental pathology detection, with the baseline experiments and code release lowering the barrier for follow-up work. The contribution is self-contained and centers on new data rather than internal parameter fitting.

major comments (2)
  1. Labeling section: the manual annotation process into the 9 classes is described but includes no quantitative validation such as inter-annotator agreement, overlap metrics, or expert review of a held-out subset. This is load-bearing for the central claim that the NN training results demonstrate dataset utility, because without evidence that the labels are accurate and consistent the reported prediction quality could simply reflect annotator-specific patterns rather than clinically meaningful boundaries.
  2. Experimental results section: the claim of 'high quality of predictions' is stated without specific metrics (e.g., mAP, mask IoU, or per-class scores), details on validation splits, or error analysis. This gap prevents full verification that the experiment supports the dataset's value for instance segmentation.
minor comments (1)
  1. Abstract: the statement that results show 'high quality of predictions' would be strengthened by including one or two concrete performance numbers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review of our manuscript on the AlphaDent dataset. We address each major comment below and have revised the manuscript to strengthen the presentation of annotation quality and experimental results where possible.

read point-by-point responses

  1. Referee: Labeling section: the manual annotation process into the 9 classes is described but includes no quantitative validation such as inter-annotator agreement, overlap metrics, or expert review of a held-out subset. This is load-bearing for the central claim that the NN training results demonstrate dataset utility, because without evidence that the labels are accurate and consistent the reported prediction quality could simply reflect annotator-specific patterns rather than clinically meaningful boundaries.

    Authors: We agree that quantitative validation of the annotations would provide stronger evidence of label reliability. The annotations were performed by a single experienced dental clinician following a detailed protocol based on standard clinical diagnostic criteria for each of the nine pathology classes. In the revised manuscript we have expanded the labeling section with additional details on the annotation guidelines, quality-control steps, and the clinician's qualifications. We did not collect multiple independent annotations for the full dataset due to resource constraints, so formal inter-annotator agreement statistics are not available; we have therefore noted this as a limitation and indicated that such metrics could be collected in future dataset extensions. revision: partial

  2. Referee: Experimental results section: the claim of 'high quality of predictions' is stated without specific metrics (e.g., mAP, mask IoU, or per-class scores), details on validation splits, or error analysis. This gap prevents full verification that the experiment supports the dataset's value for instance segmentation.

    Authors: We accept that the original phrasing was insufficiently precise. The revised manuscript now reports the concrete evaluation metrics obtained from the baseline instance-segmentation experiments, including overall mAP and mask IoU values together with per-class scores. We have also added explicit descriptions of the train/validation/test splits and a short error analysis that identifies the most frequent failure modes. These quantitative results and analyses were generated during our internal experiments and are now presented in full to allow readers to assess the dataset's utility directly. revision: yes

Circularity Check

0 steps flagged

No significant circularity: self-contained dataset contribution

full rationale

The paper introduces a new dataset AlphaDent of DSLR tooth images from 295 patients, labeled into 9 pathology classes for instance segmentation, and reports baseline neural network training results. No equations, derivations, fitted parameters, or uniqueness theorems appear. Claims rest on empirical training outcomes and open release of data/code/weights rather than any reduction to internal definitions, self-citations, or renamings. This is a standard self-contained dataset paper evaluated against external training benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on the domain assumption that the collected patient photographs and their manual labels accurately capture relevant tooth pathologies for training purposes; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption The DSLR photographs from 295 patients and their 9-class labels are accurate and representative for the instance segmentation task.
    Invoked implicitly when presenting the dataset as suitable for neural network training and reporting high-quality predictions.

pith-pipeline@v0.9.0 · 5683 in / 1299 out tokens · 54991 ms · 2026-05-19T02:59:37.275442+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages · 1 internal anchor

  1. [1]

    Publicly Available Dental Image Datasets for Artificial Intelligence,

    Uribe, S. E., Issa, J., Sohrabniya, F., Denny, A., Kim, N. N., Dayo, A. F., Chaurasia, A., Sofi -Mahmudi, A., Büttner, M., and Schwendicke, F., “Publicly Available Dental Image Datasets for Artificial Intelligence,” Journal of Dental Research 103, 1365–1374 (2024). http://doi.org/10.1177/00220345241272052

    work page doi:10.1177/00220345241272052 2024

  2. [2]

    DENTEX: An Abnormal Tooth Detection with Dental Enumeration and Diagnosis Benchmark for Panoramic X-rays,

    Hamamci, I. E., Er, S., Simsar, E., Yuksel, A. E., Gultekin, S., Ozdemir, S. D., Yang, K., Li, H. B., Pati, S., Stadlinger, B., Mehl, A., Gundogar, M., and Menze, B., “DENTEX: An Abnormal Tooth Detection with Dental Enumeration and Diagnosis Benchmark for Panoramic X-rays,” (2023). https://doi.org/10.48550/arXiv.2305.19112

    work page doi:10.48550/arxiv.2305.19112 2023

  3. [3]

    Diffusion-Based Hierarchical Multi -label Object Detection to Analyze Panoramic Dental X -Rays,

    Hamamci, I. E., Er, S., Simsar, E., Sekuboyina, A., Gundogar, M., Stadlinger, B., Mehl, A., and Menze, B., “Diffusion-Based Hierarchical Multi -label Object Detection to Analyze Panoramic Dental X -Rays,” in [ Medical Image Computing and Computer Assisted Intervention – MICCAI 2023], Greenspan, H., Mad-abhushi, A., Mousavi, P., Salcudean, S., Duncan, J., ...

    work page doi:10.1007/978-3-031-43987-2 2023

  4. [4]

    Tufts Dental Database: A Multimodal Panoramic X-Ray Dataset for Benchmarking Diagnostic Systems,

    Panetta, K., Rajendran, R., Ramesh, A., Rao, S., and Agaian, S., “Tufts Dental Database: A Multimodal Panoramic X-Ray Dataset for Benchmarking Diagnostic Systems,” IEEE Journal of Biomedical and Health Informatics 26, 1650–1659 (2022). https://doi.org/10.1109/JBHI.2021.3117575

    work page doi:10.1109/jbhi.2021.3117575 2022

  5. [5]

    A benchmark for comparison of dental radiography analysis algorithms,

    Wang, C.-W., Huang, C.-T., Lee, J.-H., Li, C.-H., Chang, S.-W., Siao, M.-J., Lai, T.-M., Ibragimov, B., Vrtovec, T., Ronneberger, O., Fischer, P., Cootes, T. F., and Lindner, C., “A benchmark for comparison of dental radiography analysis algorithms,” Medical Image Analysis 31, 63–76 (2016). https://doi.org/10.1016/j.media.2016.02.004

    work page doi:10.1016/j.media.2016.02.004 2016

  6. [6]

    AI Dental Research Dataset Catalogue

    Uribe, S. E., “AI Dental Research Dataset Catalogue.” https://github.com/sergiouribe/dental_datasets_itu/blob/ main/AI_Dental_Datasets_List.md. (Accessed: 28 July 2025)

    work page 2025

  7. [7]

    A multi - modal dental dataset for semi -supervised deep learning image segmentation,

    Wang, Y., Ye, F., Chen, Y., Wang, C., Wu, C., Xu, F., Ma, Z., Liu, Y., Zhang, Y., Cao, M., and Chen, X., “A multi - modal dental dataset for semi -supervised deep learning image segmentation,” Scientific Data 12, 117 (2025). https://doi.org/10.1038/S41597-024-04306-9

    work page doi:10.1038/s41597-024-04306-9 2025

  8. [8]

    A multimodal dental dataset facilitating machine learning research and clinic services v1.1.0

    Liu, W., Huang, Y., and Tang, S., “A multimodal dental dataset facilitating machine learning research and clinic services v1.1.0.” https://physionet.org/content/multimodal-dental-dataset/1.1.0/. (Accessed: 28 July 2025)

    work page 2025

  9. [9]

    A multimodal dental dataset facilitating machine learning research and clinic services,

    Huang, Y., Liu, W., Yao, C., Miao, X., Guan, X., Lu, X., Liang, X., Ma, L., Tang, S., Zhang, Z., and Zhan, J., “A multimodal dental dataset facilitating machine learning research and clinic services,” Scientific Data 11, 1 –11 (2024). https://doi.org/10.1038/S41597-024-04130-1

    work page doi:10.1038/s41597-024-04130-1 2024

  10. [10]

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

    Ben-Hamadou, A., Neifar, N., Rekik, A., Smaoui, O., Bouzguenda, F., Pujades, S., Boyer, E., and Ladroit, E., “Teeth3DS+: An Extended Benchmark for Intraoral 3D Scans Analysis,” arXiv:2210.06094 (2022). https://doi.org/10.48550/ARXIV.2210.06094

    work page doi:10.48550/arxiv.2210.06094 2022

  11. [11]

    SegmentAnyTooth: An open-source deep learning framework for tooth enumeration and segmentation in intraoral photos,

    Nguyen, K. D., Hoang, H. T., Doan, T. P. H., Dao, K. Q., Wang, D. H., and Hsu, M. L., “SegmentAnyTooth: An open-source deep learning framework for tooth enumeration and segmentation in intraoral photos,” Journal of Dental Sciences 20, 1110–1117 (2025). https://doi.org/10.1016/J.JDS.2025.01.003

    work page doi:10.1016/j.jds.2025.01.003 2025

  12. [12]

    GitHub: SegmentAnyTooth: An open -source deep learning framework for tooth enumeration and segmentation in intraoral photos

    Nguyen, K. D., “ GitHub: SegmentAnyTooth: An open -source deep learning framework for tooth enumeration and segmentation in intraoral photos.” https://github.com/thangngoc89/SegmentAnyTooth. (Accessed: 28 July 2025)

    work page 2025

  13. [13]

    Deep learning -based dental plaque detection on primary teeth: a comparison with clinical assessments ,

    You, W., Hao, A., Li, S., Wang, Y., and Xia, B., “Deep learning -based dental plaque detection on primary teeth: a comparison with clinical assessments ,” BMC Oral Health 20, 1 –7 (2020). https://doi.org/10.1186/S12903-020- 01114-6/

    work page doi:10.1186/s12903-020- 2020

  14. [14]

    Oral screening of dental calculus, gingivitis and dental caries through segmentation on intraoral photographic images using deep learning,

    Liu, Y., Cheng, Y., Song, Y., Cai, D., and Zhang, N., “Oral screening of dental calculus, gingivitis and dental caries through segmentation on intraoral photographic images using deep learning,” BMC Oral Health 24, 1–10 (2024). https://doi.org/10.1186/S12903-024-05072-1

    work page doi:10.1186/s12903-024-05072-1 2024

  15. [15]

    A comprehensive dataset of annotated oral cavity images for diagnosis of oral cancer and oral potentially malignant disorders,

    Piyarathne, N. S., Liyanage, S. N., Rasnayaka, R. M. S. G. K., Hettiarachchi, P. V. K. S., Devindi, G. A. I., Francis, F. B. A. H., Dissanayake, D. M. D. R., Ranasinghe, R. A. N. S., Pavithya, M. B. D., Nawinne, I. B., Ragel, R. G., and Jayasinghe, R. D., “A comprehensive dataset of annotated oral cavity images for diagnosis of oral cancer and oral potent...

    work page doi:10.1016/j.oraloncology.2024.106946 2024

  16. [16]

    Mask R-CNN

    He, K., Gkioxari, G., Dollár, P., and Girshick, R., “Mask R-CNN,” (2018). http://arxiv.org/abs/1703.06870

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  17. [17]

    Leading Image Video Data Annotation Platform — CVAT

    “Leading Image Video Data Annotation Platform — CVAT.” https://www.cvat.ai/. (Accessed: 28 July2025)

    work page

  18. [18]

    GitHub - cvat-ai/cvat: Annotate better with CVAT, the industry -leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale

    “GitHub - cvat-ai/cvat: Annotate better with CVAT, the industry -leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.” https://github.com/cvat -ai/cvat. (Accessed: 28 July 2025)

    work page 2025

  19. [19]

    M., Roberson, T

    Sturdevant, C. M., Roberson, T. M., Heymann, H. O., and Swift, E. J., [ Sturdevant’s art and science of operative dentistry], Elsevier, 7th ed. (2018)

    work page 2018

  20. [20]

    V., [ Operative dentistry: The pathology of the hard tissues of the teeth (Vol

    Black, G. V., [ Operative dentistry: The pathology of the hard tissues of the teeth (Vol. 1) ], Medico -Dental Publishing Company (1908)

    work page 1908

  21. [21]

    Kaggle: AlphaDent: Teeth marking

    “Kaggle: AlphaDent: Teeth marking.” https://www.kaggle.com/competitions/alpha-dent/. (Accessed: 30 July 2025)

    work page 2025

  22. [22]

    A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO -NAS,

    Terven, J., Córdova -Esparza, D. M., and Romero -González, J. A., “A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO -NAS,” Machine Learning and Knowledge Extraction 2023 5, 1680–1716 (2023). https://doi.org/10.3390/MAKE5040083

    work page doi:10.3390/make5040083 2023

  23. [23]

    The pascal visual object classes (VOC) challenge,

    Everingham, M., Van Gool, L., Williams, C. K., Winn, J., and Zisserman, A., “The pascal visual object classes (VOC) challenge,” International Journal of Computer Vision , 303 –338 (2010). https://doi.org/10.1007/S11263-009- 0275-4

    work page doi:10.1007/s11263-009- 2010

  24. [24]

    GitHub: AlphaDent

    “GitHub: AlphaDent.” https://github.com/ZFTurbo/AlphaDent. (Accessed: 30 July 2025)

    work page 2025