Artificial Intelligence-Based Pose Estimation Model for Anatomical Landmark Detection in Lateral Cephalometric X-Rays

In orthodontics, cephalometric analysis, which is widely used to evaluate the jaw structure, head, and facial skeleton, plays an important role in the clinical decision-making process. However, this analysis requires manual labeling of anatomical landmarks on the images of cephalometric X-rays. This process is time-consuming, prone to human error, and requires expertise. In recent years, with the prominence of deep learning and computer vision concepts, studies on automating cephalometric analysis have increased. In this study, we investigate the role of an artificial intelligence-based approach in detecting anatomical landmarks in cephalometric analysis using the YOLOv8 architecture, a recent deep learning model. The n-s-m-I submodels of the “pose” module in the YOLOv8 architecture, which vary according to the parameter values, were tested. These models, which were trained by the transfer learning technique under the problem, were compared in detail, including the model used (n-s-m-l), and hyperparameter values, before and after data augmentation. The performance criteria were evaluated, and the advantages and disadvantages of the system were discussed at the end of the study.