Key points are not available for this paper at this time.
Purpose This study aimed to evaluate the diagnostic reliability of a prototype artificial intelligence (AI)-assisted software for automated detection of clinical features from dental photographs, using sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) as metrics. Targeted features included caries, gingival recession, calculus, retained roots, bleeding, and staining. Methods A cross-sectional study was conducted on 34 patients at Ajman University's College of Dentistry, generating 306 standardized intraoral photographs. Ground truth was established by clinical examiners and compared against the prototype software output. Diagnostic performance was analyzed using Stata version 15.0 (StataCorp). Results The prototype demonstrated variable diagnostic performance. The prototype demonstrated higher sensitivity in detection of bleeding (71.43%; 95% CI: 68.74–74.11) and retained roots (72.73%; 95% CI: 66.48–78.98) with the lowest values reported for the detection of staining (24.38%; 95% CI: 21.63–27.13). The prototype demonstrated high specificity for detection of staining (98.77%; 95% CI: 98.11–99.42) and bleeding (98.88%; 95% CI: 98.15–99.45). The highest PPV was reported for the detection of staining (94.78%; 95% CI: 93.45–96.10), whereas bleeding (99.81%; 95% CI: 99.56–100.00) and retained roots (99.72%; 95% CI: 99.48–100.00) reported NPVs above 99%. Conclusion The AI software demonstrates strong specificity and NPV, indicating reliable disease exclusion. Lower sensitivity and PPV suggest the need for algorithmic refinement to enhance detection performance and reduce false positives.
Abdelaziz et al. (Fri,) studied this question.