Comparing the Effectiveness of Artificial Intelligence Technology with 6th Year Dental Students for the Diagnosis of Inflammatory Bone Lesions of the Mandible in Panoramic Radiography.
Authors
Affiliations (6)
Affiliations (6)
- Faculty of Dentistry, Khon Kaen University, Khon Kaen, Thailand. [email protected].
- Faculty of Dentistry, Khon Kaen University, Khon Kaen, Thailand.
- Department of Oral and Maxillofacial Surgery, Faculty of Dentistry, Khon Kaen University, Khon Kaen, Thailand.
- Division of Oral Diagnosis, Department of Oral Biomedical Sciences, Faculty of Dentistry, Khon Kaen University, Khon Kaen, Thailand.
- Department of Electrical Engineering, Faculty of Engineering, Khon Kaen University, Khon Kaen, Thailand.
- Department of Anatomy, Faculty of Dentistry, Mahidol University, Salaya, Thailand.
Abstract
This study aimed to evaluate the potential role of artificial intelligence (AI) as a diagnostic support tool for inexperienced clinicians by comparing its diagnostic performance and time efficiency with those of sixth-year dental students in detecting mandibular inflammatory bone lesions on panoramic radiographs. A total of 412 radiographs, taken between 2013 and 2023 from the Faculty of Dentistry, Khon Kaen University, were retrospectively collected and categorized into lesion-present (n = 192) and lesion-free (n = 220) groups, including osteomyelitis (OM), radiation-induced osteomyelitis, osteoradionecrosis, and medication-related osteonecrosis of the jaw. All images were annotated by an oral and maxillofacial radiologist and surgeon using the Roboflow platform. A You Only Look Once version 8 (YOLOv8)-based deep learning detection model was developed and evaluated using an independent test set. In parallel, 20 sixth-year dental students assessed a standardized test set of 10 panoramic radiographs (6 lesion-present and 4 lesion-free images), with diagnostic accuracy and interpretation time recorded. On an independent test set (n = 62), the AI model achieved an accuracy of 97.18% with 94.87% sensitivity and 100% specificity. When evaluated on the same standardized 10-image test set used for student comparison, the model demonstrated 90% accuracy with 100% sensitivity and 75% specificity, whereas the students achieved a mean accuracy of 29.5% (sensitivity 33.33%, specificity 23.75%; P < 0.001). The AI model also required significantly less interpretation time (0.274 s) than the students (274.05 s, P < 0.001). These findings suggest that AI demonstrates strong diagnostic capability and substantial time efficiency in detecting mandibular inflammatory bone lesions on panoramic radiographs and may serve as a valuable supportive tool to enhance diagnostic accuracy, particularly in reducing missed lesions, among less experienced clinicians.