Caries Detection on Intraoral Images Using Artificial Intelligence.

Area of Science:

• Dentistry and Dental Imaging
• Artificial Intelligence in Healthcare
• Machine Learning for Medical Diagnosis

Background:

• Visual examination (VE) is the standard for caries detection, but intraoral photographs offer machine-readable data for AI analysis.
• Current clinical practice rarely uses photographic images for diagnosis, yet they are essential for AI-driven automated analysis.
• Artificial intelligence (AI) has not been extensively applied to automatic caries detection on intraoral images.

Purpose of the Study:

• To develop a deep learning approach using convolutional neural networks (CNNs) for detecting and categorizing dental caries.
• To compare the diagnostic performance of the developed AI method against expert standards using intraoral photographs.

Main Methods:

• A dataset of 2,417 anonymized intraoral photographs of permanent teeth (occlusal and smooth surfaces) was used.
• Images were categorized into: caries-free, noncavitated caries lesion, or caries-related cavitation.
• A CNN was trained using image augmentation and transfer learning, with expert diagnoses serving as the reference standard.

Main Results:

• The CNN correctly detected caries in 92.5% of all test images (Sensitivity: 89.6%, Specificity: 94.3%, AUC: 0.964).
• When focusing on caries-related cavitation, 93.3% of tooth surfaces were correctly classified (Sensitivity: 95.7%, Specificity: 81.5%, AUC: 0.955).
• The AI method demonstrated over 90% agreement with expert standards for caries detection using standardized photographs.

Conclusions:

• A deep learning approach using CNNs can achieve high accuracy in detecting dental caries from single-tooth photographs.
• The developed AI method shows significant potential for caries detection and categorization, comparable to expert visual examination.
• Further improvements are needed for the current AI approach to be fully integrated into clinical practice.