Akademik Veri Yönetim Sistemi
Journal of Dentistry, cilt.170, 2026 (SCI-Expanded, Scopus)
Objectives: This study aimed to develop and validate an artificial intelligence (AI)-based model for quantitatively assessing gingival display during smiling using standardized extraoral photographs. The goal was to establish an objective and reproducible measurement approach that could support periodontal and esthetic evaluations. Methods: A total of 1748 standardized smiling photographs were collected, and 687 photographs showing a high smile line were used for model development. Gingival regions were manually annotated and used to train a DeepLabV3+ segmentation network with a ResNet-34 backbone. Photographs were divided into training (547), validation (70), and test (70) sets. Predicted gingival display values were compared with clinical reference measurements using root mean square error (RMSE), mean absolute error (MAE) and statistical agreement testing. Results: The model demonstrated high predictive accuracy across six gingival regions, with RMSE values of 0.62–0.90 mm and MAE values of 0.48–0.64 mm. No statistically significant differences were observed between predicted and clinical measurements (all p > 0.05). Segmentation performance was strong, with pixel accuracy of 98.4 % and intersection over union values of 99.2 % (training) and 94.3 % (validation). Conclusions: The proposed AI model provides accurate and consistent quantitative measurements of gingival display, showing close agreement with clinical assessments. These results indicate the feasibility of integrating AI-driven measurement tools into periodontal diagnostic workflows. Clinical Significance: The AI-based model provides standardized and objective millimeter-level measurements of gingival display, helping to reduce variability associated with subjective visual assessment. While time efficiency was not quantitatively assessed in this study, the proposed model may facilitate more streamlined assessment workflows. Because measurements are generated automatically, the approach may streamline clinical evaluation; however, time savings were not quantified in this study. Further research is needed to determine the extent of any efficiency benefits compared with manual assessment.