A systematic review of 26 publications on artificial intelligence models for detecting periodontal disease via intraoral photographs reveals significant accuracy variability. Classification accuracy ranged from 48% to 100%, detection from 56% to 78%, and segmentation between 43% and 70%. Convolutional neural networks were commonly used, but external validation results varied, often showing diminished performance on new populations. The review highlights limitations in study design, reporting quality, and data standards, suggesting improvements for future AI applications in periodontal diagnostics.

Source: International Dental Journal