Related Experiment Video
Updated: May 14, 2026
Induction of Periodontitis via a Combination of Ligature and Lipopolysaccharide Injection in a Rat Model
Published on: February 17, 2023
Multimodal Prediction of Periodontitis Using Root Exposure in Intraoral Images and Age
Sohee Kang1, Hyeonjeong Go2, Young-Eun Kwon3
1Department of Dentistry, College of Medicine, Yeungnam University, Daegu, Republic of Korea.
Introduction And Aims:
Despite advances in AI-based periodontitis screening, quantifiable and interpretable biomarkers from intraoral photographs remain underexplored. Therefore, this study aimed to develop a deep learning pipeline for exposed root area quantification from photographs and to evaluate its predictive value for periodontitis risk within a multimodal framework integrating age.
Methods:
Intraoral photographs of the mandibular anterior sextant and covariate questionnaires were obtained from 269 participants. A fine-tuned YOLOv11 segmentation model quantified tooth and exposed root surface areas, from which the exposed root ratio (ERR) was derived. ERR was combined with age and self-reported data to train four machine learning models (logistic regression, SVM, random forest, gradient boosting) for periodontitis prediction. Performance was assessed using AUROC and permutation feature importance across different feature sets.
Results:
The YOLOv11 segmentation model achieved an overall mAP@0.5 of 0.901, with mean Dice coefficients of 0.928 and 0.844 for tooth and exposed root, respectively. In the ≥35 age group, ERR-only models outperformed age-only models across all four machine learning algorithms, with statistically significant differences in 13 of 24 comparisons (mean ΔAUROC: 0.031-0.094, p < .05). Integration of ERR with age further improved predictive performance, yielding significant gains in 19 of 24 comparisons (mean ΔAUROC: 0.029-0.131, p < .05). Permutation feature importance analysis revealed ERR as the dominant predictor in the ≥45 age group, with importance scores of 0.391 and 0.366 for ERR compared to 0.151 and 0.273 for age in Gradient Boosting and Random Forest, respectively.
Conclusion:
AI-derived ERR from mandibular anterior images is a reproducible, interpretable biomarker that outperforms age and enhances periodontitis prediction when combined with conventional risk factors.
Clinical Relevance:
AI-driven quantification of ERR from intraoral photographs offers a practical, non-invasive, and cost-effective screening tool for periodontitis risk assessment in primary care and community settings, particularly among middle-aged and older populations.