Deep learning-based fractured tooth detection in occlusal radiographs

Area of Science:

• Dental diagnostics
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Accurate and rapid diagnosis of fractured teeth is crucial in dentistry.
• Deep learning (DL) techniques show promise for improving diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a DL-based framework for automated detection and classification of fractured teeth using occlusal radiography (OR).
• To assess the performance of various convolutional neural network (CNN) architectures and a majority voting fusion strategy.

Main Methods:

• A dataset of 200 ORs was curated for teeth 11 and 21.
• Teeth were automatically detected using a YOLOv9 framework.
• Cropped tooth images were classified using VGG19, EfficientNetB0, InceptionResNetV2, and InceptionV3 networks.
• A majority voting approach was used to fuse classification results.

Main Results:

• The YOLOv9 detector achieved high precision (mAP50: 99.5%).
• Individual CNNs achieved accuracy rates between 84.67% and 87.94%.
• The majority voting fusion improved classification accuracy to 91.94%.

Conclusions:

• The proposed DL method effectively detects fractured teeth from OR images.
• This framework pioneers the integration of AI into dental diagnostics, enhancing clinical decision-making.