A Multimodal Model for Caries Screening Using Intraoral Images and Questionnaires

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Public Health

Background:

• Dental caries is a prevalent childhood disease requiring effective screening methods.
• Current screening often relies on visual inspection, which can miss early signs.
• Deep learning offers potential for automated and accurate dental diagnostics.

Purpose of the Study:

• To develop and evaluate a multimodal deep learning model for pediatric dental caries screening.
• To integrate intraoral photographs and questionnaire data for enhanced diagnostic performance.
• To compare the multimodal model against an image-only unimodal model.

Main Methods:

• Collected 7671 intraoral images (occlusal and smooth surfaces) from children.
• Utilized oral health questionnaires alongside clinical image labeling (caries-free, early, moderate-to-severe).
• Trained and validated a multimodal deep learning model, comparing it to an image-only model using accuracy, precision, recall, F1 score, and ROC-AUC.

Main Results:

• The multimodal model demonstrated high accuracy (e.g., 92.3% for caries-free occlusal surfaces) and ROC-AUCs (e.g., 0.967 for occlusal caries-free).
• Performance for smooth surfaces was also strong, with high accuracy and ROC-AUC values.
• Integrating questionnaire data significantly improved recall for early caries without reducing overall accuracy.

Conclusions:

• The multimodal deep learning model outperforms image-only models in detecting early dental caries.
• This intelligent screening model shows potential for population-level caries surveillance.
• Further large-scale, multicenter validation is recommended to confirm generalizability.