Automated Detection and Classification of Mandibular Third Molar Impactions in Panoramic X-Rays

Mandibular third molar impaction is a common clinical issue requiring high diagnostic accuracy and efficiency due to potential complications such as infections, pain, and adj acent tooth damage. Traditional manual diagnosis is time-consuming and reliant on clinical experience, increasing the risk of misdiagnosis. To address this challenge, this paper introduces an advanced method utilizing the YOLOv7 algorithm for automatic detection and classification of mandibular third molar impactions based on Winter's classification in panoramic X-rays. Leveraging a dataset of 450 images annotated to include both second and third molars, two distinct approaches are proposed. The first approach detects and classifies both second and third molars within a single bounding box, while the second approach focuses solely on the third molar. The results showed mean average precision (mAP) scores of 0.5820 for the first approach and 0.7226 for the second approach, with a total of 10 classes (five classes on each side of the image). This method demonstrates significant potential for improving diagnostic accuracy and efficiency in dental radiography.