Automated classification of dental implant categories on panoramic radiographs using YOLOv8 and YOLO11 models

Abstract

This study investigated the feasibility of automatically classifying panoramic radiographs into four dental implant categories using YOLOv8 and YOLO11 architectures. The proposed framework aims to support automated image-level implant category classification by providing rapid image analysis together with Grad-CAM visualizations to enhance model interpretability. A publicly available dataset comprising 10,214 panoramic dental radiographs (OPGs) categorized into four implant types (Endosteal, Subperiosteal, Transosteal, and Zygomatic) was used. The dataset was divided into training, validation, and testing subsets using a stratified 70%/15%/15% split. YOLOv8 (n, s, m) and YOLO11 (n, s, m) models were trained for 50 epochs. Model performance was evaluated using precision, recall, F1-score, mAP@50, and mAP@50–95. A graphical user interface (GUI) was also developed to facilitate model inference, Grad-CAM visualization, and automated report generation. Among the YOLOv8 models, YOLOv8m achieved the best overall performance, with a test precision of 0.9837, an F1-score of 0.9790, and the highest mAP@50–95 of 0.7844. Within the YOLO11 family, YOLO11s demonstrated the most balanced performance, achieving a test precision of 0.9887, an F1-score of 0.9800, and a mAP@50–95 of 0.7665 while requiring substantially less training time than YOLOv8m. Grad-CAM visualizations qualitatively demonstrated that the models focused on radiographic regions relevant to implant category classification in representative examples. The proposed framework demonstrated promising performance for automated image-level classification of panoramic radiographs into four dental implant categories under the experimental conditions of the public dataset. Although the results indicate the potential of YOLO-based deep learning for supporting automated implant category recognition, further validation using multicenter datasets and expert-generated implant-level annotations is required before broader clinical implementation can be considered. The proposed framework may serve as a supportive tool for automated implant category classification, educational applications, research database organization, and preliminary image assessment. It is intended to complement, rather than replace, professional clinical judgment, and further validation is required to establish its performance across diverse clinical settings.

The authors' abstract, as published at the source. Scientific Reports, 2026 · DOI ↗