ArchCaries-YOLO26: a dentition-aware framework for automated detection of primary and permanent tooth caries in intraoral photographs
- 1Department of Stomatology, Zhengzhou Central Hospital Affiliated to Zhengzhou University, Zhengzhou, China
Abstract
To address the challenges of small lesion scales, irregular boundaries, substantial background interference, and similar local appearances between caries in primary and permanent teeth in color intraoral images, this paper proposes ArchCaries-YOLO26, a YOLO26-based object detection model for dental caries. Specifically, this study addresses the problem of accurately detecting and localizing carious lesions in intraoral photographs while distinguishing whether the detected lesions occur in primary or permanent teeth. The model introduces the Adaptive Dental-Arch Trajectory State-Space Module (ADT-SSM), which models the arrangement, scale, and morphological relationships of adjacent teeth through bidirectional state propagation along predicted dental-arch trajectories. Meanwhile, the Class-Conditioned Inner–Outer Boundary Integral Calibration module (CIBC) is designed to adaptively refine the bounding-box boundaries of caries in primary and permanent teeth by jointly analyzing the features of the inner candidate region, boundary ring, and outer dental tissue. On the public dataset, the proposed model achieves mAP@50, mAP@50-95, Precision, and Recall values of 0.951±0.006, 0.799±0.007, 0.918±0.005, and 0.872±0.006, respectively. On the clinical dataset, the corresponding metrics reach 0.920±0.006, 0.776±0.007, 0.779±0.009, and 0.905±0.009, respectively. The ablation results demonstrate that ADT-SSM and CIBC play complementary roles in dentition-context modeling and bounding-box localization, and their joint use achieves a more balanced performance in classification and localization. The proposed method provides a feasible technical approach for the automated screening of caries in primary and permanent teeth using conventional color intraoral images.
Related
- OsteoNet-X: A deep residual learning framework for automated osteoporosis detection from knee X-ray images
- The accuracy of large language models in assessing tooth detection and identification, dental caries, periapical status, and tooth development from mixed dentition radiographs
- Automated Regional Detection of Intervortex Venous Anastomosis on Ultra-Widefield Indocyanine Green Angiography Using a Two-Stage Deep Learning Framework
- ECG-based automated detection of sleep apnea using deep neural networks and hidden markov models
- Automated corneal endothelial cell detection and segmentation in specular microscopy images using Attention and Faster U-Net
BibTeX
@article{guo2026archcariesyolo26,
title = {{ArchCaries-YOLO26: a dentition-aware framework for automated detection of primary and permanent tooth caries in intraoral photographs}},
author = {Zhenwei Guo and Shiqian Wu and Peipei Zhang and Di Xue and Zhuo Chen},
journal = {Frontiers in Medicine},
year = {2026},
month = oct,
doi = {10.3389/fmed.2026.1955589},
url = {https://doi.org/10.3389/fmed.2026.1955589},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - ArchCaries-YOLO26: a dentition-aware framework for automated detection of primary and permanent tooth caries in intraoral photographs
AU - Zhenwei Guo
AU - Shiqian Wu
AU - Peipei Zhang
AU - Di Xue
AU - Zhuo Chen
PY - 2026
DA - 2026/10/13
JO - Frontiers in Medicine
DO - 10.3389/fmed.2026.1955589
UR - https://doi.org/10.3389/fmed.2026.1955589
AB - To address the challenges of small lesion scales, irregular boundaries, substantial background interference, and similar local appearances between caries in primary and permanent teeth in color intraoral images, this paper proposes ArchCaries-YOLO26, a YOLO26-based object detection model for dental caries. Specifically, this study addresses the problem of accurately detecting and localizing carious lesions in intraoral photographs while distinguishing whether the detected lesions occur in primary or permanent teeth. The model introduces the Adaptive Dental-Arch Trajectory State-Space Module (ADT-SSM), which models the arrangement, scale, and morphological relationships of adjacent teeth through bidirectional state propagation along predicted dental-arch trajectories. Meanwhile, the Class-Conditioned Inner–Outer Boundary Integral Calibration module (CIBC) is designed to adaptively refine the bounding-box boundaries of caries in primary and permanent teeth by jointly analyzing the features of the inner candidate region, boundary ring, and outer dental tissue. On the public dataset, the proposed model achieves mAP@50, mAP@50-95, Precision, and Recall values of 0.951±0.006, 0.799±0.007, 0.918±0.005, and 0.872±0.006, respectively. On the clinical dataset, the corresponding metrics reach 0.920±0.006, 0.776±0.007, 0.779±0.009, and 0.905±0.009, respectively. The ablation results demonstrate that ADT-SSM and CIBC play complementary roles in dentition-context modeling and bounding-box localization, and their joint use achieves a more balanced performance in classification and localization. The proposed method provides a feasible technical approach for the automated screening of caries in primary and permanent teeth using conventional color intraoral images.
SN - 2296-858X
ER -