Multi-Regional deep learning models for identifying dental restorations and prosthesis in panoramic radiographs

Area of Science:

• Artificial Intelligence in Dentistry
• Deep Learning for Medical Imaging
• Dental Radiology

Background:

• Automated detection of dental prostheses (crowns, bridges, implants) and treatments (fillings, root canals) is crucial for efficient dental practice.
• Existing methods may lack the accuracy and efficiency required for comprehensive analysis of panoramic radiographs.

Purpose of the Study:

• To develop and evaluate a novel deep learning methodology for automated detection of various dental restorations and treatments.
• To compare the performance of different deep learning architectures for this specific task.

Main Methods:

• A diverse dataset of 2,235 panoramic radiographs from three dental colleges was utilized.
• Three deep learning architectures—You Only Look Once (YOLO)v11, Faster Region-based Convolutional Neural Network (Faster R-CNN), and Vision Transformer (ViT)—were evaluated.
• Standardized protocols including dataset splitting and rigorous testing were employed.

Main Results:

• The Vision Transformer (ViT) model achieved the highest detection accuracy (94.15%), precision (94.64%), and F1-score (93.52%).
• Faster Region-based Convolutional Neural Network (Faster R-CNN) demonstrated the best mean average precision (mAP) of 82.0% at an IoU threshold of 0.50.

Conclusions:

• AI-driven automated dental restoration identification from panoramic radiographs is practical and effective.
• This technology significantly enhances diagnostic accuracy and streamlines clinical workflows in modern dental practice.
• The developed AI model offers an efficient tool for improved patient record management and clinical decision-making.