Deep Learning-Based Dental Caries Diagnosis on Panoramic Radiographies: Performance of YOLOv8 Versus Human Observers

Area of Science:

• Artificial Intelligence in Dentistry
• Deep Learning for Medical Imaging
• Radiographic Dental Diagnostics

Background:

• Dental caries detection on pediatric panoramic radiographs is crucial for early intervention.
• Deep learning models offer potential for automated analysis of radiographic images.
• Evaluating AI performance against human observers with varying experience is essential.

Purpose of the Study:

• To assess the diagnostic accuracy of a YOLOv8x deep learning model for detecting approximal, occlusal, and buccal caries in children.
• To compare the AI model's performance against dentists with different clinical experience levels.
• To determine the utility of AI as a supportive tool in pediatric dental radiography.

Main Methods:

• Retrospective analysis of 1526 pediatric panoramic radiographs (ages 5-12).
• Training and independent testing of a YOLOv8x object-detection model for caries.
• Comparison of AI performance (precision, sensitivity, F1 score) with three human observers (intern, novice specialist, experienced specialist).

Main Results:

• YOLOv8x achieved moderate performance for approximal caries (F1: 0.576) but limited for occlusal (F1: 0.24) and failed on buccal caries.
• AI performance varied by caries type, comparable to experienced specialists for approximal caries but inferior for buccal.
• Overall, the AI model's diagnostic performance was comparable to less experienced clinicians but not expert-level.

Conclusions:

• The YOLOv8x model demonstrates potential as an assistive tool for caries detection on pediatric panoramic radiographs.
• Current AI performance does not match expert-level diagnostic accuracy for all caries types.
• Further development is needed to enhance AI capabilities for comprehensive radiographic caries assessment.