Hybrid Faster R-CNN for Tooth Numbering in Periapical Radiographs Based on Fédération Dentaire Internationale System.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate tooth numbering is crucial for dental diagnosis, typically using the Fédération Dentaire Internationale system.
• Variations in periapical radiograph (PA) angles present challenges for precise tooth identification.
• Existing datasets may have limitations due to missing teeth or extensive crown loss.

Purpose of the Study:

• To develop a deep learning-based tool for accurate tooth identification and numbering in PA radiographs.
• To enhance diagnostic efficiency and accuracy in dental practices.
• To address challenges posed by variations in PA angles and data limitations.

Main Methods:

• Development of a Hybrid Faster Region-based Convolutional Neural Network (R-CNN) technique.
• Implementation of a custom loss function for PA tooth numbering to accelerate training.
• Creation of a tooth-numbering position auxiliary localization algorithm to handle missing teeth and crown loss.

Main Results:

• Achieved a maximum precision of 95.16% using a transformer-based NextViT-Faster R-CNN hybrid model.
• Demonstrated an accuracy increase of at least 8.5% compared to conventional methods.
• Reduced training time by 19.8% with the proposed auxiliary localization algorithm.

Conclusions:

• The developed AI tool effectively overcomes challenges in PA tooth numbering.
• The method enhances clinical efficiency and reduces the risk of misdiagnosis in AI-assisted dental diagnostics.
• The findings support the integration of advanced AI tools in routine dental practices.