Detection of pulpal calcifications on bite-wing radiographs using deep learning

Area of Science:

• Artificial Intelligence in Dentistry
• Radiographic Image Analysis
• Deep Learning Algorithms

Background:

• Pulpal calcifications are calcified masses within the dental pulp cavity.
• Accurate detection of these calcifications is crucial for dental diagnosis.
• Current detection methods can be time-consuming and subjective.

Purpose of the Study:

• To evaluate the performance of the YOLOv4 deep learning algorithm for automated detection of pulpal calcifications.
• To assess the algorithm's ability to identify calcification in pulp chambers on bite-wing radiographs.

Main Methods:

• Trained the YOLOv4 algorithm using 2000 labeled bite-wing radiographs (80% training, 10% validation, 10% testing).
• Utilized transfer learning to obtain optimal weight files for pulpal calcification detection.
• Evaluated the algorithm's performance on unseen test radiographs, with results verified by oral radiologists.

Main Results:

• The YOLOv4 algorithm achieved high performance in detecting pulp chambers (recall 86.98%, precision 98.94%, F1-score 91.60%).
• Detection of pulpal calcification showed strong results: recall 86.39%, precision 85.23%, F1-score 85.49%, and accuracy 96.54%.
• The algorithm demonstrated high success rates in identifying both pulp chambers and calcifications.

Conclusions:

• The YOLOv4 algorithm effectively detects pulp chambers and calcifications in bite-wing radiographs.
• This deep learning approach offers a highly accurate decision support system for dentists in clinical practice.