Developing deep learning methods for classification of teeth in dental panoramic radiography

Area of Science:

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

Background:

• Diagnostic interpretation errors and time constraints can impact dental treatment effectiveness.
• Artificial intelligence (AI) offers potential solutions for improving dental diagnostic accuracy and efficiency.
• Deep learning models are increasingly being explored for complex image analysis tasks in healthcare.

Purpose of the Study:

• To develop an AI-based clinical dental decision-support system using deep learning.
• To compare the performance of You Only Look Once V4 (YOLO-V4) and Faster Regions with the Convolutional Neural Networks (R-CNN) for tooth classification.
• To determine the most effective deep learning method for accuracy, speed, and detection ability in dental panoramic radiography.

Main Methods:

• Retrospective analysis of 1200 dental panoramic radiographs.
• Training deep learning models on a semantic segmentation task for tooth classification.
• Comparison of YOLO-V4 and Faster R-CNN performance metrics including precision, recall, and F1 score.

Main Results:

• YOLO-V4 achieved 99.90% precision, 99.18% recall, and 99.54% F1 score.
• Faster R-CNN achieved 93.67% precision, 90.79% recall, and 92.21% F1 score.
• YOLO-V4 demonstrated superior accuracy, speed, and detection of impacted and erupted third molars compared to Faster R-CNN.

Conclusions:

• The YOLO-V4 deep learning method is more effective than Faster R-CNN for tooth classification in dental panoramic radiography.
• AI-driven decision-support systems can assist dentists, saving time and reducing the impact of stress and fatigue.
• The proposed deep learning approach enhances clinical decision-making in daily dental practice.