Comparison of dental age estimation from panoramic radiographs using deep learning-based artificial intelligence applications and the demirjian method

Comparison of Dental Age Estimation from Panoramic Radiographs Using Deep Learning-Based Artificial Intelligence Applications and the Demirjian Method Aim: The aim of this study is to evaluate the performance of a deep learning-based artificial intelligence (AI) model, developed using tooth development stages labeled according to the Demirjian method on pediatric panoramic radiographs, in estimating dental age. Additionally, the study aims to compare the model's accuracy and reliability with those of expert and researcher observers who applied the same method manually. Materials and Methods: In this retrospective descriptive study, 1,910 panoramic radiographs of children aged 5 to 14 were evaluated. The mandibular left permanent teeth were manually labeled according to the Demirjian method, and the data were used to train an AI model developed with YOLOv8x-based CranioCatch software. The model's performance in segmentation, object detection, and classification was analyzed. Dental age estimations were calculated by a pediatric dentistry specialist, a research assistant, and the AI model, and statistically compared. Results: The AI model demonstrated a moderate level of success in classifying dental development stages. While no significant difference was found between methods in female subjects, the dental ages estimated from the stages labeled by the AI and the researcher were significantly higher than the chronological age in male subjects (p<0,05). Overall, the AI model tended to produce higher dental age values compared to the human observers. A high level of inter-observer agreement was observed among all evaluators (ICC=0,927). Conclusion: According to the results of this study, although the AI model based on the Demirjian method showed an average level of variation in dental age estimation, a high level of inter-observer agreement was found across all methods. This study suggests that deep learning algorithms have potential as supportive tools in dental age estimation; however, further studies are needed to enhance model accuracy and improve class-specific performance. Keywords: Artificial intelligence, Deep learning, Demirjian method, Dental age estimation, Pediatric dentistry, YOLOv8x