Deep‑learning‑based detection of open‑apex teeth on panoramic radiographs using YOLO models

Area of Science:

• Dentistry
• Artificial Intelligence
• Radiology

Background:

• Detecting teeth with open apices is crucial for treatment planning.
• Conventional methods may require multiple radiographs, increasing patient radiation exposure.
• Deep learning offers a potential solution for efficient and accurate tooth apex detection.

Purpose of the Study:

• To evaluate and compare the performance of You Only Look Once (YOLO)-based deep learning models for detecting open-apex teeth.
• To assess the diagnostic accuracy of YOLOv3, YOLOv4, and YOLOv5 in identifying open-apex teeth on panoramic radiographs.

Main Methods:

• Utilized 966 panoramic radiographs, with open-apex teeth labeled into six classes (incisors, premolars, molars).
• Trained and evaluated YOLOv3, YOLOv4, and YOLOv5 artificial intelligence models.
• Calculated precision, recall, mean average precision (mAP), and F1 score to assess model performance.

Main Results:

• YOLOv4 demonstrated the highest overall performance with a mean average precision (mAP@0.5) of 87.84%.
• YOLOv4 achieved the highest recall (90%) and F1 score (0.87).
• All evaluated YOLO models (v3, v4, v5) exceeded 84% in key performance metrics.

Conclusions:

• YOLO-based deep learning models show significant promise for detecting open-apex teeth.
• These AI systems can provide faster, more accurate diagnoses, potentially reducing the need for additional radiographs.
• The findings suggest deep learning can aid dentists by improving diagnostic efficiency and patient care.