AI-POWERED CARIES DETECTION IN PRIMARY TEETH USING DEEP LEARNING ANALYSIS OF DIGITAL RADIOGRAPHS AND INTRAORAL IMAGES

Background: Dental caries in primary teeth remains a significant global health challenge, with early detection critical
for preventing complications. Traditional diagnostic methods rely on subjective visual-tactile examination and
radiography, which exhibit limitations in sensitivity and reproducibility. Artificial intelligence (AI) offers promising
solutions for objective caries detection.
Objective: To develop and validate a deep learning (DL) model for detecting caries in primary teeth using digital
radiographs and intraoral images, and compare its performance against expert dentists.
Methods: A retrospective study included 600 digital radiographs (bitewing) and 600 intraoral images from 320
children (aged 3–8 years). A convolutional neural network (CNN) was trained on 70% of the data, validated on 15%,
and tested on 15%. Ground truth was established by two pediatric dentists. Performance metrics (accuracy, sensitivity,
specificity) were calculated and compared to three general dentists using McNemar’s test.
Key Findings: The DL model achieved an accuracy of 94.3% ± 1.4 for radiographs and 92.7% ± 1.6 for intraoral
images. Sensitivity and specificity were 93.5% and 95.1% for radiographs, and 91.8% and 93.6% for intraoral images,
respectively. The model significantly outperformed general dentists (accuracy: 84.2% ± 3.1; p < 0.001) and matched
pediatric dentists (accuracy: 93.8% ± 1.2; p = 0.21).
Conclusion: The DL model demonstrates high diagnostic accuracy for caries detection in primary teeth, surpassing
general dentists and offering a reliable adjunctive tool for clinical practice.