Annotated intraoral image dataset for dental caries detection.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Lack of publicly available annotated intraoral image datasets hinders AI development for dental caries detection.
• Dental caries is a prevalent global health issue requiring efficient diagnostic tools.

Purpose of the Study:

• To introduce the first publicly available annotated intraoral image dataset for AI-driven dental caries detection.
• To establish a benchmark for AI model performance in detecting dental caries.

Main Methods:

• Collected 6,313 intraoral images from individuals aged 10-24 in Pakistan.
• Annotated images using LabelMe, verified by dentists, and converted to YOLO, PASCAL VOC, and COCO formats.
• Trained and evaluated five AI models (YOLOv5s, YOLOv8s, YOLOv11, SSD-MobileNet-v2, Faster R-CNN).

Main Results:

• The dataset includes diverse intraoral views of mixed and permanent dentitions.
• YOLOv8s demonstrated the best performance with a mean Average Precision (mAP) of 0.841 at 0.5 Intersection over Union (IoU).

Conclusions:

• The developed dataset and AI model benchmarks significantly advance AI-based dental diagnostics for caries detection.
• Future research should incorporate primary dentition and varied imaging devices.