Feasibility of multi-class dental caries detection using deep learning-based smartphone images: a pilot prospective...

Area of Science:

• Artificial Intelligence in Dentistry
• Digital Health Technologies
• Oral Health Diagnostics

Background:

• Dental caries detection traditionally relies on visual examination, which can be subjective.
• Smartphone imaging offers a accessible platform for potential oral health monitoring.
• Deep learning models show promise in analyzing medical images for diagnostic support.

Purpose of the Study:

• To evaluate the feasibility of using deep learning models with smartphone images for multi-class dental caries detection.
• To assess the performance of a YOLOv6-L6 model trained on self-taken intraoral images.
• To explore the potential of smartphone-based AI as a tool for patient-driven oral health screening.

Main Methods:

• Seventy adults provided self-taken intraoral smartphone images.
• Images were annotated by dentists using binary, four-class, and five-class caries severity scales (ICDAS-II, ADA CCS).
• A YOLOv6-L6 deep learning model was trained and evaluated using metrics like sensitivity, specificity, precision, accuracy, F1-score, and mAP.

Main Results:

• The deep learning models achieved high performance across all classification schemes (sensitivity ≥86%, specificity >97%, accuracy >93%).
• F1-scores remained high (>88%), with a slight decrease as classification complexity increased.
• The binary classification model yielded the highest mean average precision (mAP) at 65.2%, while the five-class model achieved the highest specificity (98.45%).

Conclusions:

• Smartphone-based deep learning is a feasible approach for dental caries detection.
• This technology can serve as a supportive tool for dentists and facilitate patient-driven screening.
• Further research can optimize AI models for more nuanced caries classification.