Background: Recent advancements in artificial intelligence (AI) have substantially enhanced diagnostic precision in dentistry, particularly in detecting radiopaque lesions on panoramic radiographs. However, clinical integration remains limited due to anatomical overlap, dataset heterogeneity, and insufficient external validation. Objectives: This scoping review aimed to systematically identify and evaluate the diagnostic performance of AI-based models for detecting radiopaque lesions on panoramic radiographs, emphasizing metrics such as accuracy, sensitivity, specificity, precision, F1-score, and area under the curve (AUC). Methods: Following the PRISMA-ScR framework, a comprehensive search was conducted across PubMed, ScienceDirect, Wiley Online Library, and grey literature for studies published between 2015 and 2025. Eligible English-language studies evaluated AI algorithms for radiopaque lesion detection with quantitative performance data. Five studies involving 22,529 participants met the inclusion criteria. Extracted data covered AI architectures, lesion types, and diagnostic outcomes. Results: Among the identified models, InceptionResNetV2, YOLOv5, YOLOv8, and RCNN demonstrated superior diagnostic performance. InceptionResNetV2 achieved up to 96% accuracy, YOLOv5 attained 98.1% precision, and YOLOv8 yielded the highest F1-score (98.12%) and accuracy (98%) for osteoporosis detection. Conclusion: AI models—particularly InceptionResNetV2, YOLOv5, and YOLOv8—exhibit remarkable diagnostic potential in detecting radiopaque lesions on panoramic radiographs. Nevertheless, dataset variability and limited multicenter validation highlight the need for standardized methodologies to ensure consistent clinical applicability.
Barung et al. (Thu,) studied this question.