Deep Learning-Based Detection and Anatomical Classification of Retained Roots and Periodontally Compromised Teeth in Orthopantomogram across Multi-National Populations

The purpose of this study was to develop and evaluate deep learning models for the automated detection of retained roots in panoramic radiographs, with a specific focus on comparing oriented bounding
box (OBB) and axis-aligned bounding box (AABB) annotation strategies across diverse populations.A multinational dataset of 4,768 panoramic radiographs was annotated into seven clinically
relevant classes using both oriented bounding boxes (OBB) and axis-aligned bounding boxes (AABB) approaches. You Only Look Once version 11 (YOLOv11) and Real-Time Detection Transformer (RT-DETR) models were trained under identical conditions and evaluated using mean average precision (mAP) and loss metrics. Inference results were further examined through qualitative case analysis to capture strengths and
limitations in clinically challenging scenarios.