Paper Title
AI-Driven Comparative Analysis of YOLO-Based Models for Multi-Class Detection in Panoramic Radiographs

Abstract
Automated Detection of multiple types of dental pathology in radiographic images via deep learning methods is still largely a single-class task, which does not take into account the practical complexity of the problem. In this work, we propose anapproach to dental pathology detection with panoramic images in four classes: caries, fillings, impacted tooth, and implant. We have developed a consistent pipeline of annotation based on normalization of CSV coordinates in the YOLO format with use of the letterbox augmentation technique and class-balancing loss. The three generations of YOLO (YOLOv5, YOLOv7, and YOLOv8) neural network architectures were examined using equal training conditions. YOLOv8 showed the highest overall performance in terms of accuracy (mAP@50 = 0.782, mAP@50-95 = 0.520). YOLOv7 showed the highest recall (0.759), whereas YOLOv5 demonstrated the best precision (0.774). Detecting cavities appeared to be the hardest task for all considered YOLO architectures. This framework establishes a reproducible baseline for future clinical computer-aided diagnosis integration. Keywords - Dental Pathology Detection; Panoramic Dental Radiographs; Deep Learning; Multi-Class Object Detection; Real-Time Object Detection Networks; Computer-Aided Dental Diagnosis; Artificial Intelligence in Dentistry.