Tooth Detection and Numbering Performance of Two AI-Based Systems on Panoramic Radiographs: A Comparative Study

Objectives: To evaluate and compare the tooth detection and numbering performance of two commercially available artificial intelligence (AI) software systems, ThakaaMed and EM2AI, in the automated detection and numbering of permanent teeth on panoramic radiographs. Methods: A retrospective study was conducted using a randomized sample of 595 panoramic radiographs obtained from multiple dental outpatient facilities. The sample included patients aged 6–18 years with mixed or permanent dentition. Ground truth (GT) annotations were established independently by three experienced dental clinicians using the FDI tooth numbering system. Each radiograph was analyzed using ThakaaMed Dental IQ (Version 1.9) and EM2AI (Version 3.2.0), and AI-generated outputs were compared with GT annotations. Tooth detection and numbering performance was assessed using sensitivity, specificity, and accuracy, and sensitivities and specificities were compared between the two AI systems across anatomical regions. Third molars were analyzed separately because of their variable developmental stages in mixed dentition radiographs. Results: EM2AI demonstrated superior performance in third molar detection compared with ThakaaMed. Excluding third molars, overall diagnostic accuracy was 96.81% for ThakaaMed and 99.48% for EM2AI. Regional analyses demonstrated significantly higher sensitivity for EM2AI in the upper anterior, lower anterior, and lower posterior regions, whereas no significant differences in specificity were observed between the two AI systems. EM2AI maintained consistently high tooth detection and numbering performance across anatomical regions. Conclusions: Both ThakaaMed and EM2AI demonstrated favorable tooth detection and numbering performance and potential utility as supportive tools in routine dental diagnostic and documentation workflows. EM2AI demonstrated greater consistency across anatomical regions and superior performance in third molar detection. These findings highlight the importance of evaluating AI systems across diverse anatomical regions and developmental stages when assessing their clinical applicability.