Diagnostic Accuracy of Artificial Intelligence-Assisted Intraoral Imaging for Detection of Early Dental Caries: A Multicenter Clinical Study

Abstract

Background: Early detection of dental caries may prevent lesion progression and facilitate minimally invasive treatment. This study aimed to assess the diagnostic performance of artificial intelligence (AI) supported intraoral imaging in the detection of early dental caries in adults. Methods: This multicenter clinical study included 480 adults from three dental centers. A total of 1,920 tooth surfaces were assessed using AI-assisted intraoral imaging and conventional clinical assessment. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, F1 score, Cohen's κ, and area under the receiver operating characteristic curve (AUC) were evaluated. Results: The reference standard identified 520 (27.1%) surfaces with early dental caries. AI-assisted intraoral imaging showed 91.3% sensitivity (95% CI: 88.6–93.5%), 89.2% specificity (95% CI: 87.5–90.7%), 75.9% PPV, 96.5% NPV, and 89.8% overall accuracy. The AUC was 0.943 (95% CI: 0.929–0.956) and Cohen's κ was 0.747, suggesting good agreement with the reference tests. The sensitivity was highest for lesions involving dentin (96.4%) and was satisfactory for early enamel lesions (87.6%). AI-assisted intraoral imaging had a significantly higher sensitivity (91.3% vs. 76.5%) and AUC (0.943 vs. 0.851; both P<0.001) than conventional clinical evaluation. The mean diagnostic interpretation time was also lower when using AI-assisted intraoral imaging (0.42 ± 0.11 vs. 0.68 ± 0.17 min/surface; P<0.001). Conclusion: AI-assisted intraoral imaging showed high diagnostic accuracy and good agreement with the reference standard for early dental caries detection. Further prospective validation in different populations and clinical settings is warranted before routine widespread use.