Development and validation of artificial intelligence models for automated periodontitis staging and grading using panoramic radiographs.

Abstract

Background: Periodontal diseases are common chronic conditions that can lead to tooth loss and systemic complications if not diagnosed and treated promptly. The 2017 classification by the American Academy of Periodontology highlights the need for effective, accurate diagnostic tools. This study aimed to develop and validate an AI-driven system for automated staging and grading periodontitis from panoramic radiographs using the YOLOv8 architecture.

Methods: A total of five hundred panoramic radiographs from patients diagnosed with periodontitis were included. Radiographs were labeled and split into training (75%), validation (15%), and testing (10%) sets. Three specialized YOLOv8-based models were trained to segment the alveolar bone level, the cemento-enamel junction (CEJ), and tooth axes. Image augmentations were applied to enhance model robustness. The resulting measurements of radiographic bone loss were combined with patient information (age, smoking status, diabetes) to identify periodontitis stage and grade following the 2017 guidelines.

Results: The bone level and CEJ detection models achieved high precision (0.95-0.97) and recall (0.94-0.96), reflecting strong segmentation performance. The tooth detection model achieved a precision of approximately 0.82 and a recall of 0.81. Integrating all three models enabled automated determination of periodontal stage (I-IV) and grade (A-C), with an interactive interface allowing clinicians to review and adjust outputs if necessary.

Conclusion: The proposed YOLOv8-based framework accurately detects key periodontal landmarks and automates disease staging and grading. Future work should expand the dataset, refine the tooth detection model, and validate the system in clinical settings to support large-scale periodontal screening and improved patient care.