Diagnostic accuracy of an artificial intelligence-based software in detecting supernumerary and congenitally missing teeth in panoramic radiographs

Abstract

Background/Objectives: Recent advances in AI have enabled its application in dentistry. This study assessed the diagnostic accuracy of an AI-based model (Diagnocat™) in detecting congenitally missing and supernumerary teeth on panoramic radiographs.

Materials/Methods: Three groups of 50 orthopantomograms each—control, congenitally missing, and supernumerary teeth—were evaluated by two human observers and Diagnocat™. Diagnostic performance was compared using the Wilcoxon Signed Rank and McNemar’s tests. Agreement was measured using Cohen’s Kappa, and diagnostic metrics (sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV)) were computed using IBM SPSS 29.0.

Results: For congenitally missing teeth, Cohen’s Kappa indicated strong agreement (0.91); however, significant differences were found in the diagnostic performance (p < 0.01). The model exhibited 84.7% sensitivity, 100.0% specificity, 100.0% PPV, and 99.4% NPV. For supernumerary teeth, the agreement was moderate (Kappa = 0.60), with significant differences in the diagnostic performance (p < 0.001). Sensitivity was 43.9%, while specificity, PPV, and NPV were 100.0%, 100.0%, and 98.9%, respectively.

Limitations: Using convenience sampling and a retrospective design may affect generalizability and applicability.

Conclusions/Implications: Although the AI-based model shows promise, it is not yet able to replace human assessment as the standard for detecting missing and supernumerary teeth in panoramic radiographs