Application of artificial intelligence in predicting periodontal risks in orthodontic patients: a scoping review

Introduction This scoping review aims to evaluate artificial intelligence (AI), particularly deep learning, for predicting periodontal risks in orthodontic patients and to identify methodological gaps limiting clinical translation. Methods Reported in accordance with PRISMA-ScR and PRISMA-S guidelines, a systematic search was conducted across six electronic databases (PubMed, IEEE Xplore, Web of Science, Scopus, Embase, Cochrane Library), grey literature, and citation tracking (Jan 1, 2018–June 1, 2025). Out of 880 identified records (312 duplicates removed; 568 screened), 56 full texts were evaluated, yielding 3 core clinical CBCT studies meeting strict eligibility criteria (Xu et al., 2024; Pirayesh et al., 2024; Zheng et al., 2025). Methodological quality and risk of bias were appraised using QUADAS-2 and PROBAST. Results Reported performance for orthodontically induced inflammatory root resorption (OIIRR) ranged from 82.1% accuracy in small canine-induced resorption cohorts to >94% accuracy in larger grading frameworks. However, critical appraisal revealed a high risk of bias across all studies due to single-center retrospective designs, lack of independent external validation, and missing calibration metrics. Direct clinical evidence for AI prediction of gingival recession or orthodontic alveolar bone loss was insufficient within strict pre-cutoff parameters. Discussion While AI demonstrates technical feasibility for assessing orthodontically related root resorption risk, robust individual-level risk prediction remains unestablished. Prospective multicenter external validation and adherence to standardized reporting guidelines are required prior to routine clinical adoption.