Development of an artificial intelligence model for automated detection of the mandibular canal, mental foramen, and maxillary sinus in panoramic radiographs: an original diagnostic accuracy study.
Authors
Abstract
The objective of this study was to train, create, and test an artificial intelligence model for automated detection of mandibular canal, mental foramen, and maxillary sinus in panoramic radiographs to evaluate the model's reliability and potential for integration into routine practice. The study utilized 10,000 panoramic radiographs each carefully annotated to mark the mandibular canal, mental foramen, and maxillary sinus. These images were divided into training and tuning datasets in a 4:1 ratio to train the model, which was subsequently tested on a novel set of 636 panoramic radiographs. The model's performance was assessed by categorizing the results into true positives, true negatives, false positives, and false negatives. Based on these outcomes, key performance metrics like sensitivity, specificity, accuracy, and precision were calculated to evaluate the model's diagnostic effectiveness and clinical applicability. The model achieved 100.0% sensitivity for the mandibular canal and maxillary sinus, while the mental foramen showed 93.9% sensitivity and 94.8% specificity. Precision was highest for the mandibular canal (100.0%), followed by the mental foramen (99.7%) and maxillary sinus (99.5%). Accuracy was 100.0% for the mandibular canal, 99.5% for the maxillary sinus, and 93.9% for the mental foramen. The study concluded that the model effectively segments anatomical landmarks like mandibular canal, mental foramen, and maxillary sinus in panoramic radiographs and can be successfully integrated with dental radiographic workflows. With clinical integration and continued validation, these tools could improve diagnostic consistency, and aid in decision-making in general dental practice.