Scholarlink

<p>This study aims to estimate the gonial angle from dental panoramic radiographs using deep learning models with transfer learning and keypoint detection. A dataset of 1000 panoramic X-rays collected between 2023 and 2024 was used. Three anatomical landmarks—Articulare, Gonion, and Menton (Ar, Go, Me)—essential for calculating the gonial angle were annotated by an oral radiologist. The Individual Keypoint Labeling (IKL) method was applied, labeling each point separately with three bounding boxes. The dataset was divided into training (80%), validation (10%), and testing (10%) sets. Two deep learning models, YOLOv8 and YOLO11, were trained using the IKL method and transfer learning. Performance was evaluated using spatial and angular error metrics, including mean absolute error (MAE), root mean square error (RMSE), and accuracy metrics such as precision, recall, <i>F</i>1-score, and mean average precision (mAP). Prediction reliability was also assessed using Bland–Altman analysis and Intraclass Correlation Coefficient (ICC). The YOLO11 model showed superior results in both spatial and angular accuracy. On the test data, it achieved a pose <i>F</i>1-score of 0.9975, mAP@50 of 0.995, and mAP@50:95 of 0.992. In terms of angular accuracy, it yielded an MAE of 2.37° and an RMSE of 2.58°, outperforming the YOLOv8 model. These findings demonstrate that the YOLO11 model provides a more accurate and clinically reliable approach for automatic gonial angle estimation in panoramic radiographs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Gonial Angle Estimation on Dental Panoramic X-Rays Using Deep Learning: A Transfer Learning Approach for Keypoint Detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月7日

查看全文
  • Muhammet Üsame Öziç,
  • Serap Akdoğan,
  • Melek Tassoker

摘要

This study aims to estimate the gonial angle from dental panoramic radiographs using deep learning models with transfer learning and keypoint detection. A dataset of 1000 panoramic X-rays collected between 2023 and 2024 was used. Three anatomical landmarks—Articulare, Gonion, and Menton (Ar, Go, Me)—essential for calculating the gonial angle were annotated by an oral radiologist. The Individual Keypoint Labeling (IKL) method was applied, labeling each point separately with three bounding boxes. The dataset was divided into training (80%), validation (10%), and testing (10%) sets. Two deep learning models, YOLOv8 and YOLO11, were trained using the IKL method and transfer learning. Performance was evaluated using spatial and angular error metrics, including mean absolute error (MAE), root mean square error (RMSE), and accuracy metrics such as precision, recall, F1-score, and mean average precision (mAP). Prediction reliability was also assessed using Bland–Altman analysis and Intraclass Correlation Coefficient (ICC). The YOLO11 model showed superior results in both spatial and angular accuracy. On the test data, it achieved a pose F1-score of 0.9975, mAP@50 of 0.995, and mAP@50:95 of 0.992. In terms of angular accuracy, it yielded an MAE of 2.37° and an RMSE of 2.58°, outperforming the YOLOv8 model. These findings demonstrate that the YOLO11 model provides a more accurate and clinically reliable approach for automatic gonial angle estimation in panoramic radiographs.