Automated Tooth Numbering and Periodontal Bone Loss Severity Detection Using Residual U-Net and Tooth Long Axis in Panoramic Radiography

Abstract

Periodontitis is a major cause of tooth loss, and its assessment from panoramic radiographs remains subjective. This study proposes an integrated framework for automatic tooth numbering and periodontal bone loss severity assessment using a residual U-Net architecture. The framework performs cemento-enamel junction (CEJ) segmentation, alveolar bone crest (ABC) segmentation, and tooth numbering based on the Fédération Dentaire Internationale (FDI) system. It estimates tooth long-axis using an oriented bounding box (OBB) for anatomically consistent bone loss measurement. Several residual backbones are evaluated, with ResNet34 achieving the best performance. Results obtain dice scores of 0.968 and 0.960 for CEJ and ABC segmentation, respectively, and 0.852 for tooth numbering. Bone loss severity evaluation on 265 teeth from 10 panoramic radiographs shows that the OBB-based approach outperforms axis-aligned bounding boxes (AABB) method, improving accuracy from 0.61 to 0.63 and macro-F1 score from 0.45 to 0.47. These results demonstrate the feasibility of automated and anatomically consistent periodontal assessment.