Artificial Intelligence for the Detection of the Palatoglossal Airspace in Panoramic Radiographs: A Diagnostic Support Tool

OBJECTIVE: To develop and validate an artificial intelligence model capable of identifying the presence of the palatoglossal airway in panoramic radiographs, a factor that may impair diagnostic image quality, thereby providing an intermediate technical evaluation tool prior to diagnostic image analysis. MATERIALS AND METHODS: A total of 456 panoramic radiographs were selected from a radiographic database containing approximately 10,000 images and independently classified by three evaluators into control (tongue positioned on the palate) and test (tongue not positioned on the palate) groups, used to establish the gold standard for comparison with the algorithm's classification. The AI model was developed using YOLOv11n architecture detection for training, validation, and testing, respectively. The model's performance was evaluated using the metrics of sensitivity, specificity, accuracy, recall, precision, and F1 score. RESULTS: The AI model achieved 89.3% accuracy in detecting the presence of a palatoglossal airway associated with the classification of compromised image quality, demonstrating balanced sensitivity and specificity (both 89.47%) and excellent agreement with the reference standard (Kappa index = 1.0). CONCLUSION: The proposed AI-based model demonstrated high accuracy and robustness in detecting tongue-to-palate malposition in panoramic radiographs, a frequent error associated with the palatoglossal airway space that compromises diagnostic image quality. CLINICAL RELEVANCE: The developed algorithm is available through the website, where it is possible to obtain an instant technical analysis of the panoramic image, capable of preventing errors in subsequent diagnostic interpretation.