Umeå universitets logga

OBJECTIVE: Detecting marginal jaw bone loss on radiographs is crucial for diagnosing periodontitis but remains difficult and time-consuming. This review evaluated artificial intelligence (AI) accuracy in identifying the alveolar bone crest and estimating bone loss compared with dental professionals. Moreover, we also assessed whether AI models can detect changes in bone levels over time.

METHODS: We conducted a systematic review in accordance with the PRISMA guidelines, with diagnostic accuracy as the primary outcome. The review protocol was registered in PROSPERO (CRD42024517330). Searches were performed in PubMed, Web of Science, Cochrane, and Scopus up to August 2025. Two independent reviewers screened the articles at the abstract and title levels, and performed full-text and risk-of-bias assessments. A qualitative synthesis was complemented by a random-effects meta-analysis of studies reporting binary classification of marginal bone loss.

RESULTS: Sixty-four studies met the inclusion criteria, with 16 included in the meta-analysis. AI models demonstrated promising performance in detecting the alveolar bone crest and showed high diagnostic accuracy for marginal bone loss, with a pooled sensitivity of 92.3%, a specificity of 91.7%, and an AUC of 0.97. However, high heterogeneity and frequent risk of bias were identified. No study evaluated changes in bone levels over time or was performed in a clinical setting.

CONCLUSION: AI holds promise for facilitating diagnostic decision-making in periodontal care. However, its clinical utility remains limited due to methodological issues. Future research should emphasize external validation, diverse datasets, and longitudinal image analysis to better align AI tools with real-world diagnostic needs.