AI-aided diagnosis of periodontitis in oral X-ray images

Abstract

This study explored the construction and application of efficient deep-learning models to assist the diagnosis of periodontitis in panoramic radiographs. A periodontitis auxiliary diagnosis dataset was constructed in collaboration with the Peking University School of Stomatology. The dataset included 238 panoramic images, covering different stages of healthy teeth and periodontitis. The Labelme annotation tool was used to label tooth instances, alveolar bone contours, and the cemento-enamel junction. A Mask R-CNN model was developed for tooth segmentation, and a U-Net model was developed for segmenting alveolar bone contours and cemento-enamel junctions. Based on the results of tooth instance segmentation, principal component analysis was utilized to fit the direction of the dental long axis. The minimal bounding rectangle of the tooth prediction mask was used to determine the length of the tooth axis. The proportion of alveolar bone loss was calculated based on the distance of the cemento-enamel junction and the alveolar bone level along the dental long axis. An evaluation was conducted on 20 panoramic images comprising 496 teeth. The study achieved an accuracy rate of 90.73% in the staging of periodontitis.