Journal of clinical periodontology · 2026
Can Artificial Intelligence Match Human Expertise in Long-Term Periodontal Prognosis? A Comparative Accuracy Study.
СтатьяСравнительное исследованиеНазвание 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.
López-Galindo Pablo Antonio, Troiano Giuseppe, Serroni Matteo, Dias Debora R, Santamaría Pasquale, Nibali Luigi, Avila-Ortiz Gustavo, Ravidà Andrea
Аннотация
AIM: To compare the prognostic performance of an artificial intelligence (AI) model with that of experienced clinicians in predicting tooth loss over a 10-year period.
MATERIALS AND METHODS: An AI model trained on structured clinical and radiographic data was compared with 12 periodontists and 11 general dentists (GDs), who independently assigned prognostic scores (0-10 scale) to 300 teeth with known 10-year outcomes. AI and clinician performance were evaluated under a fixed threshold and group-specific optimal thresholds. Accuracy, sensitivity, specificity, predictive values, area under the receiver operating characteristic curve and calibration were used for comparison.
RESULTS: Using the fixed threshold (score > 5 = survival), clinicians achieved higher overall accuracy than AI (75.6% periodontists, 74.9% GDs, 69.2% AI; p < 0.05), with sensitivity low and comparable across groups (14.7%-22.7%). Using group-specific optimal thresholds, AI achieved the highest accuracy for all-cause tooth loss (67.5%) and performed comparably to GDs for periodontitis-related tooth loss (76.2% vs. 76.3%). Calibration was substantially worse for AI (Brier score 0.247) than for periodontists and GDs (0.171, 0.170). Individually, AI performance overlapped with that of several clinicians.
CONCLUSION: Clinicians outperformed AI under a fixed threshold, but AI performance was comparable to clinicians using optimal thresholds, situating AI within the range of experienced clinicians despite modest overall predictive ability.
MATERIALS AND METHODS: An AI model trained on structured clinical and radiographic data was compared with 12 periodontists and 11 general dentists (GDs), who independently assigned prognostic scores (0-10 scale) to 300 teeth with known 10-year outcomes. AI and clinician performance were evaluated under a fixed threshold and group-specific optimal thresholds. Accuracy, sensitivity, specificity, predictive values, area under the receiver operating characteristic curve and calibration were used for comparison.
RESULTS: Using the fixed threshold (score > 5 = survival), clinicians achieved higher overall accuracy than AI (75.6% periodontists, 74.9% GDs, 69.2% AI; p < 0.05), with sensitivity low and comparable across groups (14.7%-22.7%). Using group-specific optimal thresholds, AI achieved the highest accuracy for all-cause tooth loss (67.5%) and performed comparably to GDs for periodontitis-related tooth loss (76.2% vs. 76.3%). Calibration was substantially worse for AI (Brier score 0.247) than for periodontists and GDs (0.171, 0.170). Individually, AI performance overlapped with that of several clinicians.
CONCLUSION: Clinicians outperformed AI under a fixed threshold, but AI performance was comparable to clinicians using optimal thresholds, situating AI within the range of experienced clinicians despite modest overall predictive ability.