구강 내 사진과 치아 모형 스캔을 이용한 교정 발치 패턴 예측에서의 인공지능 정확도

다국어 초록 (Multilingual Abstract)

Introduction: This study aimed to evaluate the accuracy of an artificial intelligence model in predicting orthodontic extraction patterns using intraoral photographs and dental model scans. Methods: Orthodontic treatment plans and pre-treatment patient data, intra-oral (IO) images and 300 right-side digital scans of pre-treatment gypsum model casts, were collected from patients who had completed orthodontic treatment at the Department of Orthodontics. Data set consisted of 150 extraction cases and 150 non-extraction cases, divided into 85% training and 15% test set for AI models. Model performance was evaluated by accuracy, precision, recall, and F1 scores. Results: The AI model trained with IO images at level 2 showed 65.90% accuracy, 66.80% precision, 65.90% recall, and 66.30% F1 score. In addition, the AI model trained with the digital scans at level 1 showed 61.63% accuracy, 61.60% precision, 61.40% recall, and 61.50% F1 score. At level 2, the digital scan model showed 61.18% accuracy, 68.30% precision, 68.20% recall, and 68.20% F1 score. Moreover, the AI model trained with the digital scans at level 3 showed 43.18% accuracy, 41.80% precision, 43.20% recall, and 42.50% F1 score. Among the tested models, the AI model trained with the digital scan level 2 showed the highest scores. Conclusions: These findings indicate the potential utility of an artificial intelligence model in supporting extraction decision-making in orthodontic treatment based on intraoral photographs or digital model scans.