Automated detection and classification of dental trauma in periapical radiographs using deep learning: a study based on the Andreasen classification.

OBJECTIVE

The aim of this study was to evaluate the performance of a deep learning (DL) model in automatically identifying dental trauma types on selected periapical radiographs, classified according to the Andreasen system.

METHODS AND MATERIALS: Selected periapical radiographs were annotated based on the Andreasen classification. Using these annotations, a YOLOv8-based DL model was developed to classify trauma types. Because of the large number of trauma subtypes and the limited dataset size, labels were later consolidated into two main categories, and a second model was trained. The performance of both models was assessed using sensitivity, precision, and F1-score.

RESULTS

The initial model showed low overall performance, with a sensitivity of 0.34, precision of 0.29, and F1-score of 0.31. Among the subtypes, avulsion achieved the best performance across all metrics (F1-score: 0.83). After regrouping labels into two main categories, the model's overall performance improved markedly (F1-score: 0.76). Performance was higher for detecting "injuries to hard dental tissues and the pulp" (F1-score: 0.82) than for "injuries to the periodontal tissues" (F1-score: 0.44).

CONCLUSION

The DL model demonstrated strong potential in identifying dental trauma on selected periapical radiographs, particularly in accurately localizing fracture lines.