AI Plaque Scoring From Disclosed Photos Matched Agreement Between Two Examiners

Edited by Hygienetown staff

A deep-learning system that measured plaque on disclosed intraoral photographs agreed with a calibrated dentist’s plaque scores as closely as two human examiners agreed with each other, according to a study published online Oct. 19 in the Journal of Dentistry.

Researchers photographed 20 adults in one frontal and two lateral views after 48 hours without oral hygiene. A calibrated dentist who was blinded to the model scored the same teeth using the Turesky-modified Quigley-Hein plaque index. Two deep-learning models, one for each type of view, calculated the stained percentage of each outlined buccal crown.

Across 438 teeth, automated plaque coverage ranked teeth in close agreement with the clinical index, with a Spearman’s ? of 0.856. After investigators calibrated the thresholds used to convert coverage into index grades, agreement with the dentist reached a weighted ? of 0.812, compared with 0.791 between two human examiners. Automated grades fell within one grade of the clinical score for 96.8% of teeth.

The calibration step mattered. Using the index’s literal area thresholds, the system over-graded plaque by 1.16 grades, and agreement fell to a weighted ? of 0.460.

At the patient level, mean automated and clinical scores correlated at r = 0.930, with an intraclass correlation coefficient of 0.916.

Eduarda Freitas de Andrade, Karolina Felisberto, Robert Lucas Bonifácio Correia, Jamil Awad Shibli, and Mohamed A. Hassan of the Department of Periodontology at Universidade Guarulhos in Brazil conducted the study. It was supported by the State of São Paulo Research Foundation, and the authors declared no competing interests.

The authors said automated grading could serve as a second opinion that supports rather than replaces a clinician’s plaque scoring. The sample was small and the analysis covered buccal surfaces only, and the authors said the system still needs external validation.

Sources:
Journal of Dentistry, “A deep-learning system for dental plaque quantification from disclosed intraoral photographs, validation against clinical plaque scoring,” published online Oct. 19, 2026: sciencedirect.com/pii/S0300571226007931