Deep Learning-Based Multi-Tooth Segmentation on Panoramic Radiographs Using YOLOv8 Architecture

Abstract

This research introduces a multi-class tooth-level segmentation framework on panoramic radiographs using YOLOv8, trained on clinically annotated Indonesian dental data. A dataset of 302 annotated panoramic radiographs from patients at Universitas Andalas Dental Hospital was utilized, with each tooth precisely labeled according to international dental nomenclature. The model was trained using transfer learning with the YOLOv8 variant, optimized with the Adam algorithm, and evaluated using precision, recall, F1-score, and Intersection over Union (IoU). The results demonstrate that YOLOv8 is not only effective for lesion detection but also robust for fine-grained anatomical dental segmentation. The performance achieved 93.72% accuracy, 92.67% precision, 98.88% recall, and 95.58% F1-score, indicating high accuracy in tooth detection and boundary delineation. Qualitative analysis confirmed accurate segmentation across a wide range of anatomical variations, including crowding, impaction, and prosthetics. This research establishes YOLOv8 as a highly effective tool for dental image segmentation, offering significant potential to improve diagnostic efficiency, support odontological forensics, and enable automated patient record management. Future work will focus on integrating multi-class pathology detection and 3D reconstruction.