Optimized Multi-Scale Detection and Numbering of Teeth in Panoramic Radiographs Using DentifyNet

Abstract

Manual tooth detection and numbering in panoramic radiographs are time-consuming and prone to human errors, negatively impacting diagnostic accuracy and treatment outcomes. These challenges necessitate robust automated solutions to improve efficiency and precision in dental imaging. This study introduces DentifyNet, a novel deep learning architecture designed for automatic tooth detection and numbering in panoramic radiography images. DentifyNet integrates a customized Faster R-CNN framework with Feature Pyramid Networks (FPN), flexible anchor structures, and RoI Align to enhance detection precision. The model was trained and evaluated on 468 panoramic radiographs annotated by dental experts using the FDI numbering system. Experimental results demonstrate that DentifyNet achieved state-of-the-art performance with an Intersection over Union (IoU) of 86.42%, precision of 97.52%, recall of 97.49%, F1-score of 97.51%, and accuracy of 97.50%. The architecture effectively detects challenging cases, such as adjacent similar teeth and missing teeth. These findings suggest that DentifyNet surpasses standard Faster R-CNN architectures, offering a reliable solution for automated tooth detection and numbering. Future research will focus on utilizing broader datasets and architectural advancements to address current limitations and expand clinical applications.