• Introduction
• News
• PRAD Dataset
• Model
• Pending Work
• Citation

⭐️ PRAD is a large-scale periapical radiograph segmentation dataset, annotated and verified by two endodontists with over 10 years of clinical experience. It provides nine anatomical-level segmentation labels, including:

• Tooth
• Alveolar Bone (AB)
• Pulp
• Root Canal Filling (RCF)
• Denture Crown (DC)
• Dental Fillings (DF)
• Implant (IM)
• Orthodontic Devices (OD)
• Apical Periodontitis (AP)

⭐️ PRAD is designed to provide a valuable supplement to the dental deep learning community, addressing the data gap in periapical radiographs and establishing benchmark models to foster the thriving development of this field.

• [2026-05-29] PRAD++, an upgraded version of PRAD, has been accepted by IEEE TMI！
• [2025-09-21] The codes of benchmark model PRNet has been released!
• [2025-09-20] The PRAD dataset is now officially open for public application. You are welcome to use it!
• [2025-08-30] All review and revision work for the PRAD dataset has been completed. Congratulations!
• [2025-08-12] Our logo of PRAD has been confirmed!
• [2025-06-17] Our paper has been accepted by MICCAI 2025 as a poster paper! See you in Daejeon!
• [2025-04-10] Our paper has been uploaded to arXiv.
• For any questions or you need help, please seed a email to aics@nankai.edu.cn or zzh_nkcs@mail.nankai.edu.cn

Due to copyright and commercial considerations, the PRAD dataset will be partially available to the public through an application process. If you wish to use the PRAD dataset, please follow the procedure below to submit the application form for our review and processing:

• Please download the application form PRAD_application.pdf from this repository, complete it with your basic information, affiliated institution, and the purpose of using PRAD, and provide an official seal or signature.
• Please compose an email with the subject line PRAD-application+Name+Institution, attach the completed application form, send it to aics@nankai.edu.cn and copy to zzh_nkcs@mail.nankai.edu.cn.
• Once you have sent the email, your application will formally enter the review process. We will process your request as soon as possible, with an estimated completion time of 14 working days, so please plan accordingly.
• Upon approval, we will reply to your email with a download link and usage instructions for the PRAD dataset. You will then be granted access to 5,000 images and their corresponding segmentation labels from PRAD.

PRNet is a benchmark segmentation model specifically designed for the PRAD dataset task.

To use PRNet, locate the class PRNet(nn.Module): in the model.py file of the repository, modify the model parameters as needed, and then easily integrate it into your training code for training and evaluation.

• 🔲 MM-PRAD: A multimodal PRAD dataset with textual descriptions is currently under development. (Expected to be released later！)
• 🔲 Developed PRNet++ to perform end-to-end segmentation and classification tasks on PRAD. (Coming soon!🔥🔥)
• 🔲 Release the multi-label classification annotations of PRAD. (Coming soon!🔥🔥)
• ✅ Release the codes of PRNet.
• ✅ Release the PRAD Dataset.
• ✅ Release the codes of PRNet.
• ✅ Fully review, modification and verification of the dataset again.

If you find our work helpful, please star this repository and cite our paper:

@inproceedings{zhou2025prad,
  title={PRAD: Periapical Radiograph Analysis Dataset and Benchmark Model Development},
  author={Zhou, Zhenhuan and Zhang, Yuchen and Xu, Ruihong and Zhao, Xuansen and Li, Tao},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={475--484},
  year={2025},
  organization={Springer}
}

Copyright © College of Computer Science, Nankai University. All rights reserved.