Artificial Intelligence Applications in the Diagnosis and Management of Periodontitis: A Review

Résumé fourni par la source

Abstract Periodontitis, a major global public health burden, remains a leading cause of tooth loss in adults. Recent advances in artificial intelligence (AI), particularly machine learning and deep learning, show substantial potential to enhance diagnostic accuracy, treatment planning, and prediction of disease progression. This systematic review demonstrates AI applications across periodontal diagnosis, classification, prognosis prediction, treatment planning, and patient monitoring; appraised methodological quality; and identified gaps for future research and clinical translation. This review was conducted in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses 2020 statement. Six major databases, PubMed/MEDLINE, Scopus, Web of Science, EMBASE, Cochrane Library, and Google Scholar, were searched from January 1, 2010 to June 6, 2025. Predefined eligibility criteria selected original peer-reviewed studies applying AI, machine learning, or deep learning to periodontal disease and reporting at least one quantitative performance metric. Two reviewers independently screened, extracted, and appraised studies. Risk of bias was assessed using a domain-based tool adapted from QUADAS-2 and PROBAST. Due to substantial heterogeneity, results were synthesized narratively with descriptive statistics for comparable metrics; no meta-analysis was performed. Twenty-eight studies met the inclusion criteria. Convolutional neural networks and hybrid architectures dominated the imaging-based literature, with diagnostic accuracies ranging from 73.0 to 99.4% (median ≈91%) for radiographic detection of alveolar bone loss. Classification and staging studies reported accuracies ranging from 70 to 98% (median ≈87%), with consistently higher performance in advanced stages than in early-stage disease. Prognosis prediction models reported accuracies between 73 and 95% (median ≈80%), with probabilistic graphical models integrating clinical and salivary biomarker data reaching area under the receiver operating characteristic curve values up to 0.88. Risk of bias was high or unclear in most studies, primarily due to retrospective designs, internal-only validation, and imbalanced data sets. AI demonstrates performance comparable to, and sometimes exceeding, that of human experts in well-defined diagnostic tasks, particularly in radiographic detection of alveolar bone loss. However, evidence is constrained by heterogeneity and limited external validation. Current findings support AI as an adjunctive decision-support tool rather than a substitute for clinical judgment. Prospective multicenter studies with standardized reporting and external validation are required for safe clinical translation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.