Machine learning assisted classification of periodontal health and periodontitis using alveolar bone loss measurements on bitewing radiographs

Background Periodontitis is a prevalent inflammatory disease characterized by progressive loss of periodontal attachment and alveolar bone. Conventional diagnostic approaches, including periodontal probing and radiographic interpretation, are influenced by examiner variability, limiting consistency in large-scale and community-based assessments. Machine learning-driven models may support standardized screening for periodontal health and periodontitis. Aim To develop and validate a machine learning algorithm to classify periodontal health and periodontitis using clinician-measured and validated radiographic alveolar bone loss obtained from bitewing radiographs, aligned with the 2018 American Academy of Periodontology/European Federation of Periodontology (AAP/EFP) classification of periodontal and peri-implant diseases and conditions. Methods In this retrospective study, 1,537 of 2,162 bitewing radiographs were included. Alveolar bone loss was measured from the cemento-enamel junction to the alveolar crest using MiPACS software and validated by calibrated examiners ( κ =0.94). Data were preprocessed, balanced, and split into training, validation, and test sets. Multiple classifiers were benchmarked, with Random Forest (RF) selected after hyperparameter tuning. Performance was assessed using accuracy, sensitivity, specificity, F1 score, and area under the receiver operating characteristic curve (AUC). Results The RF model achieved 96.4% accuracy on the validation dataset and 92% on the independent, previously unseen test dataset. Sensitivity was 100% for periodontitis and 93% for healthy cases in the validation set, with an AUC of 0.99. On the unseen test dataset, sensitivity was 94% for healthy and 92% for periodontitis, with an AUC of 0.97. Conclusions The machine learning algorithm accurately classified periodontal health and periodontitis from bitewing radiographs, providing an automated assessment aligned with the 2018 AAP/EFP classification. It may improve diagnostic consistency, support early intervention, and enable community-based screening and triage for large-scale periodontal assessment. Clinical relevance Automated classification of periodontal status from bitewing radiographs may improve diagnostic consistency and facilitate efficient screening and triage in large-scale and community-based dental care.