Effect of data size on tooth numbering performance via artificial intelligence using panoramic radiographs

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate tooth numbering is crucial for dental diagnosis and treatment planning.
• Automated detection of tooth numbering problems using artificial intelligence can enhance diagnostic efficiency.

Purpose of the Study:

• To investigate the impact of dataset size on the performance of deep learning models for detecting tooth numbering issues in dental panoramic radiographs.
• To determine the optimal amount of data required for robust tooth numbering model training.

Main Methods:

• Utilized a dataset of 3000 anonymous adult dental panoramic X-rays, labeled according to the FDI tooth numbering system.
• Trained the YOLOv4 algorithm on subsets of 1000, 1500, 2000, and 2500 images.
• Evaluated model performance using metrics like F1 score, mAP, sensitivity, and precision on a fixed test set of 500 images.

Main Results:

• Model performance consistently improved with an increase in the number of training data.
• The model trained with 2500 images demonstrated the highest accuracy and reliability in tooth numbering detection.
• Key performance indicators (F1 score, mAP, sensitivity, precision) showed a positive correlation with dataset size.

Conclusions:

• Dataset size is a critical factor influencing the accuracy of automated dental tooth enumeration.
• Larger sample sizes are essential for developing more reliable and effective AI-driven dental diagnostic tools.
• Findings support the use of extensive datasets for training deep learning models in dental imaging applications.