Journal of periodontal research · 2026
Artificial Intelligence in Periodontology: A Systematic Review.
СтатьяСистематический обзорНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Tichy Antonin, Werner Nils, Dujic Helena, Wetzel Charlotte, Pitchika Vinay, Bumm Caspar Victor, Folwaczny Matthias, Schwendicke Falk
Аннотация
AIM: To provide a comprehensive review of artificial intelligence (AI) applications in periodontology, focusing (1) on deep learning for image-based diagnosis of periodontitis and (2) on non-image-based AI applications across periodontal care.
METHODS: This study adhered to PRISMA guidance. Six databases (PubMed, Scopus, Web of Science, Embase/Ovid, IEEE Xplore, and arXiv) were searched. The first review question (PICO 1) focused on applications of deep learning to human imaging data for diagnosing periodontitis, and the systematic review was followed by a modified QUADAS-2 risk-of-bias (RoB) assessment. The second part (PICO 2) scoped AI applications in periodontology using non-imaging data. Because of substantial heterogeneity in tasks, inputs, and outcomes, PICO 2 was synthesized narratively without formal RoB assessment.
RESULTS: PICO 1 included 29 studies, predominantly using panoramic radiographs (n = 21). Binary periodontitis classification achieved accuracies of 81%-99% on panoramic radiographs and 78% on CBCT, whereas staging/severity showed lower performance (accuracy 64%-91% in panoramic radiographs; 83% in intraoral radiographs with AUROC 0.84-0.93). Photograph-based screening achieved AUROC 0.93. RoB was generally low, but applicability concerns were frequent, mainly because of single-center datasets. PICO 2 included 65 studies, covering diagnosis and classification of periodontitis (AUROC 0.77-0.85), risk stratification and screening (AUROC 0.60-0.98), progression, and treatment outcome modeling (AUROC 0.58-0.89), oral-systemic associations, biomarker identification, and clinical data mining using natural language processing, which achieved near-perfect metrics.
CONCLUSION: Generalizability remains the key limitation across applications, driven by limited data diversity, inconsistent tasks/metrics, and scarce external testing. Future studies should prioritize multicenter evaluation, transparent reporting, and prospective assessments of workflow impact and patient-related outcomes.
REGISTRATION: PROSPERO identification number CRD420251128758.
METHODS: This study adhered to PRISMA guidance. Six databases (PubMed, Scopus, Web of Science, Embase/Ovid, IEEE Xplore, and arXiv) were searched. The first review question (PICO 1) focused on applications of deep learning to human imaging data for diagnosing periodontitis, and the systematic review was followed by a modified QUADAS-2 risk-of-bias (RoB) assessment. The second part (PICO 2) scoped AI applications in periodontology using non-imaging data. Because of substantial heterogeneity in tasks, inputs, and outcomes, PICO 2 was synthesized narratively without formal RoB assessment.
RESULTS: PICO 1 included 29 studies, predominantly using panoramic radiographs (n = 21). Binary periodontitis classification achieved accuracies of 81%-99% on panoramic radiographs and 78% on CBCT, whereas staging/severity showed lower performance (accuracy 64%-91% in panoramic radiographs; 83% in intraoral radiographs with AUROC 0.84-0.93). Photograph-based screening achieved AUROC 0.93. RoB was generally low, but applicability concerns were frequent, mainly because of single-center datasets. PICO 2 included 65 studies, covering diagnosis and classification of periodontitis (AUROC 0.77-0.85), risk stratification and screening (AUROC 0.60-0.98), progression, and treatment outcome modeling (AUROC 0.58-0.89), oral-systemic associations, biomarker identification, and clinical data mining using natural language processing, which achieved near-perfect metrics.
CONCLUSION: Generalizability remains the key limitation across applications, driven by limited data diversity, inconsistent tasks/metrics, and scarce external testing. Future studies should prioritize multicenter evaluation, transparent reporting, and prospective assessments of workflow impact and patient-related outcomes.
REGISTRATION: PROSPERO identification number CRD420251128758.