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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oral and Maxillofacial Surgery

Background:

• The gonial angle is a crucial metric in cephalometric analysis, often manually measured from dental panoramic radiographs.
• Manual measurement can be time-consuming and subject to inter-observer variability.
• Automated methods are needed to improve efficiency and consistency in gonial angle estimation.

Purpose of the Study:

• To develop and evaluate deep learning models for automatic gonial angle estimation from panoramic radiographs.
• To compare the performance of YOLOv8 and YOLO11 models using transfer learning and keypoint detection.
• To assess the clinical reliability of the proposed automated method.

Main Methods:

• A dataset of 1000 panoramic X-rays was utilized, with key anatomical landmarks (Articulare, Gonion, Menton) annotated by an oral radiologist.
• The Individual Keypoint Labeling (IKL) method was employed for annotation.
• Two deep learning models, YOLOv8 and YOLO11, were trained using transfer learning and evaluated with spatial and angular error metrics.

Main Results:

• The YOLO11 model demonstrated superior performance compared to YOLOv8 in estimating the gonial angle.
• YOLO11 achieved a pose F1-score of 0.9975 and mAP@50:95 of 0.992 on the test dataset.
• Angular accuracy for YOLO11 was high, with a Mean Absolute Error (MAE) of 2.37° and Root Mean Square Error (RMSE) of 2.58°.

Conclusions:

• Deep learning models, particularly YOLO11, offer a highly accurate and reliable method for automatic gonial angle estimation from panoramic radiographs.
• This automated approach has the potential to enhance diagnostic efficiency and consistency in oral and maxillofacial radiology.
• The findings support the clinical applicability of AI in quantitative cephalometric analysis.