Abstract Objectives To assess the diagnostic accuracy of a commercial artificial intelligence system for automated tooth numbering on panoramic radiographs and cone-beam computed tomography and to quantify case-level reliability. Methods In this retrospective single-centre diagnostic accuracy study, consecutive patients who underwent both panoramic radiography and cone-beam computed tomography in 2024 were included. The index test was automated tooth numbering generated by Diagnocat using the Fédération Dentaire Internationale numbering scheme. The reference standard was modality-specific consensus of two experienced, blinded readers. Tooth-position performance metrics with 95% confidence intervals were estimated using patient-level cluster bootstrap. Case-level reliability was defined as the proportion of examinations with completely error-free numbering across all evaluated tooth positions. Results The study analysed 178 panoramic radiographs and 174 cone-beam computed tomography examinations. Tooth-position performance was near-perfect and similar across modalities (overall accuracy 99.79% for panoramic radiographs and 99.80% for cone-beam computed tomography). At the case level, 154/178 (86.5%) panoramic radiographs and 156/174 (89.7%) cone-beam computed tomography examinations were error-free. Conclusions Despite near-perfect tooth-position metrics, approximately one in ten to one in seven examinations required at least one manual correction, demonstrating a gap between granular accuracy and case-level reliability. Advances in knowledge Reporting case-level, error-free outputs alongside tooth-position metrics provides a more clinically meaningful estimate of reliability for automated tooth numbering and supports safer workflow implementation.
Sultani et al. (2026) studied this question.