ABSTRACT

Objective

Accurate detection of mental foramen in panoramic radiological images is of great importance for numerous clinical and diagnostic applications in dentistry.

Methods

The present study has aimed to investigate the performance of multiple versions of the YOLO family (YOLOv5, YOLOv8, YOLOv9, YOLOv10, YOLOv11 and YOLO World) and advanced deep learning-based object detection models, including the RT-DETR architecture, in the identification and localization of mental foramen. A dataset of 100 anonymized panoramic radiological images manually annotated by an oral maxillofacial surgery specialist was used for model training and evaluation.

Results

The models were evaluated using standard metrics such as precision, sensitivity and mAP50 and the RT-DETR-x architecture achieved the highest performance with 89.0% precision, 85.0% sensitivity and 85.6% mAP50. In particular, YOLO World-s and YOLOv9e provided competitive results by effectively balancing detection accuracy and computational efficiency. The findings highlight the clinical applicability of AI-based detection systems in automating and facilitating dental diagnosis, while demonstrating how robust deep learning models are in handling complex anatomical variations and imaging challenges.

Conclusion

This research provides a solid foundation for future advances in automated dental diagnosis, with recommendations for expanding the dataset and integrating advanced ensemble techniques to  improve model generalization and accuracy.