Diagnostic accuracy of artificial intelligence for detecting periodontal bone loss on dental radiographs: a systematic review

Introduction: Periodontitis is staged in part by radiographic alveolar bone loss, whose manual measurement is subjective and time-consuming. Artificial intelligence has been proposed to automate this measurement, but reported accuracy varies widely. Methods: Following PRISMA 2020, PubMed, OpenAlex, and Crossref were searched for studies published from 2018 to 2025 that evaluated artificial intelligence models for detecting, staging, or measuring periodontal bone loss on panoramic or periapical radiographs against a reference standard. Risk of bias was assessed with QUADAS-2. Results: Thirty-eight studies were included, most using panoramic radiographs and convolutional neural networks. Reported sensitivity ranged from 0% to 100%, specificity from 66.2% to 98%, and the area under the curve from 0.70 to 0.951. Staging and measurement studies reported accuracy from 58% to 96.8%. Conclusions: Artificial intelligence detects periodontal bone loss with high but heterogeneous accuracy. The evidence is strongest for detection on panoramic radiographs and weakest for severity staging and clinical measurement, and external validation remains uncommon.