A Hybrid YOLOv8s+Swin-T Transformer Approach for Automated Caries Detection on Periapical Radiographs

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Early detection of dental caries is crucial for effective treatment.
• Current deep learning models show variable accuracy on dental X-ray datasets, limiting their clinical use.
• Automating caries detection in intraoral radiographs is needed.

Purpose of the Study:

• To develop and evaluate a robust hybrid object detection model for automated dental caries detection.
• To improve the accuracy and adaptability of deep learning models for caries detection in intraoral periapical radiographs.

Main Methods:

• A hybrid object detector combining Swin-T transformer and YOLOv8s backbone was developed.
• The model was trained on 1887 intraoral periapical radiographs.
• Hierarchical attention mechanism was employed for enhanced feature extraction.

Main Results:

• The proposed YOLOv8s+Swin-T model achieved high performance metrics: 0.97 for precision, recall, and F1-score.
• The model demonstrated superior performance compared to YOLOv8 variants and Faster R-CNN.
• Achieved an mAP@0.5 of 0.99, indicating excellent detection accuracy.

Conclusions:

• The YOLOv8s+Swin-T model shows significant clinical applicability and robustness for dental caries detection.
• This AI tool provides a reliable method for accurate caries detection.
• Supports routine AI-assisted diagnosis in dentistry.