一种卷积神经网络与视觉变换器在生前全景牙科X光片中的人体识别比较研究
A comparative study of convolutional neural networks and vision transformers for human identification from ante-mortem panoramic dental radiographs
该研究评估了卷积神经网络和视觉变换器在生前全景牙科X光片中的人体识别性能。ConvNeXt-S在验证任务中表现最佳,准确率为0.9908;ConvNeXt-L在识别任务中表现最佳,Rank@1和Rank@5准确率分别为0.9877和0.9983。研究结果表明,深度度量学习在法医牙科识别中有潜在应用价值。
- Leonardo Antonio Moros Morales1,
- Céline Sabourin3,5,
- Caroline Dunoyer1,2,
- Eric Baccino3,4 &
- …
- Camille Inquimbert2,5
We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply.
Abstract
Identifying deceased individuals is a critical component of forensic investigations, particularly following mass casualty events or in cases involving advanced decomposition, where traditional identification methods such as fingerprint analysis may no longer be applicable. Dental records provide a valuable alternative; however, their comparison remains a challenging and time-consuming task for forensic experts. In this study, we investigate the use of deep learning for human identification from panoramic dental radiographs. We employ triplet loss to learn discriminative image embeddings and evaluate six modern neural network architectures, including convolutional and transformer-based models, using panoramic radiographs collected at Montpellier University Hospital (CHU Montpellier). Performance was assessed on two tasks: verification, which determines whether two images belong to the same individual, and identification, which consists of retrieving the identity of an unknown image from a database of known individuals. Among the evaluated architectures, the best verification performance was achieved by ConvNeXt-S, with an accuracy of 0.9908. For identification, the best-performing model was ConvNeXt-L, achieving a Rank@1 accuracy of 0.9877 and a Rank@5 accuracy of 0.9983. These results demonstrate the potential of deep metric learning for forensic dental identification and suggest that semi-automatic systems could substantially assist practitioners during the identification process.
Acknowledgements
We thank the Clinical Data Warehouse (Entrepôt de Données de Santé eDOL) of the Montpellier University Hospital for its support and the realization of data management and data curation tasks. We thank Jorge Galicia and Valérie Macioce for their assistance with proofreading the manuscript.
Funding
This work was supported by the Montpellier University Hospital, France.
Author information
Authors and Affiliations
Health Data Science Unit, Public Health Service, University Hospital of Montpellier, Montpellier, France
Leonardo Antonio Moros Morales & Caroline Dunoyer
Desbrest Institute of Epidemiology and Public Health, UMR UA11 University of Montpellier - INSERM, Montpellier, France
Caroline Dunoyer & Camille Inquimbert
EDPFM, UR-UM212, University of Montpellier, Montpellier, France
Céline Sabourin & Eric Baccino
Faculté de Médecine de Montpellier, University of Montpellier, Montpellier, France
Eric Baccino
Dental Department, Montpellier University Hospital, University of Montpellier, Montpellier, France
Céline Sabourin & Camille Inquimbert
Authors
- Leonardo Antonio Moros Morales
- Céline Sabourin
- Caroline Dunoyer
- Eric Baccino
- Camille Inquimbert
Corresponding author
Correspondence to Leonardo Antonio Moros Morales.
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Moros Morales, L.A., Sabourin, C., Dunoyer, C. et al. A comparative study of convolutional neural networks and vision transformers for human identification from ante-mortem panoramic dental radiographs. Sci Rep (2026). https://doi.org/10.1038/s41598-026-79974-1
Received: 02 September 2026
Accepted: 04 November 2026
Published: 18 November 2026
DOI: https://doi.org/10.1038/s41598-026-79974-1
来源:Nature 健康护理主题 · nature.com