Detection Performance of an Artificial Intelligence Model in Identifying Anatomical Landmarks on Dental Panoramic Radiographs

Abstract

Manual identification of panoramic radiographs has limitations due to distortion, superimposition, and individual anatomical variations, so more objective supporting technology such as artificial intelligence (AI) is needed. The purpose of this study was to determine the level of AI detection performance in determining anatomical landmarks in panoramic radiographs. This descriptive study used a purposive sampling method. A total sample of 653 radiographs was used. The radiographs were labeled at nine anatomical landmarks and then analyzed using the YOLOv8 deep learning algorithm. The level of AI detection performance was evaluated using precision, sensitivity (recall), and F1-Score parameters in percentage form. The results showed variations in the level of AI detection performance for each anatomical landmark. The highest detection performance was in the condyloid process, while the mental foramen showed the lowest detection performance. In conclusion, AI shows a high level of detection performance in determining anatomical structures in the jaw and this can be used to assist clinicians in detecting the presence of these anatomical structures.