Deep learning detection of ectopic canines and molars in mixed dentition

Ectopic eruption (EE) during the mixed dentition is a significant dental anomaly that requires early detection to prevent complications that may impact function, aesthetics, and health. Advances in artificial intelligence point to the need for automated methods to detect ectopic canines and molars. This study aimed to develop a DL model to automatically detect EE of canines and molars on panoramic radiographs. This retrospective study utilised a dataset of panoramic radiographs from pediatric patients in the mixed dentition stage. EE of canines and molars was defined using angular and positional criteria. Images were annotated by calibrated specialists and divided into training, validation, and test sets. Performance was evaluated using precision, recall, F1-score, Dice coefficient, and mean average precision (mAP). For ectopic canines, precision was 0.786, recall 0.771, F1-score 0.778, Dice coefficient 0.768, and mAP 0.793. For ectopic molars, precision was 0.812, recall 0.650, F1-score 0.722, Dice coefficient 0.757, and mAP 0.719. These results indicate more consistent detection performance for ectopic canines than for molars. DL demonstrated effective diagnostic capability for detecting EE on panoramic radiographs. This tool has potential to improve early diagnosis, support pediatric dental treatment planning, and enhance radiology education.