Deep learning-based assessment of pulp involvement in primary molars using YOLO v8

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Dental caries is a significant global health issue, particularly in children.
• Accurate assessment of pulp involvement is critical for effective pulp treatment in pediatric dentistry.
• Current diagnostic methods for pulp involvement lack significant improvement.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for diagnosing pulp involvement in carious primary molars using radiographs.
• To assess the model's diagnostic performance, particularly differentiating between upper and lower primary molars.

Main Methods:

• A convolutional neural network (CNN) model, YOLOv8m-cls, was trained on 482 bitewing radiographs from children's dental records.
• Images were preprocessed, standardized, and augmented for classification of pulp involvement.
• Model performance was evaluated using top-1 and top-5 accuracy, precision, recall, and F1-scores.

Main Results:

• The YOLOv8m-cls model achieved a top-1 accuracy of 78.7% for upper primary molars and 87.8% for lower primary molars.
• Higher accuracy was observed for lower primary teeth in validation datasets.
• The model demonstrated effectiveness in diagnosing pulp involvement on new test images.

Conclusions:

• A promising CNN model was developed for diagnosing pulp involvement in primary teeth from bitewing radiographs.
• The model shows potential for clinical application in pediatric dentistry, aiding diagnosis and treatment planning.
• Future research should focus on integrating clinical variables and utilizing whole bitewing images for enhanced diagnostic capabilities.