Application of Deep Learning for Automated Detection of Dental Caries and Periodontal Diseases in Radiographic Images

Dental caries (DC) and periodontal disease (PD) affect nearly ۳.۵ billion people globally, presenting significant health and economic challenges. Traditional radiographic evaluation remains the standard diagnostic approach, yet it is constrained by two-dimensional projections, anatomical overlapping, and considerable intra- and inter-observer variability among clinicians. Recent developments in deep learning (DL), including convolutional neural networks (CNNs), vision transformers, and single-stage object detectors such as YOLO variants, offer promising solutions to overcome these diagnostic limitations. This review synthesizes current evidence on DL methodologies applied to intraoral and extraoral dental radiography—specifically bitewing, periapical, and panoramic imaging. DL models demonstrate high diagnostic accuracy, frequently exceeding ۹۰% sensitivity and specificity for caries localization and automated staging of alveolar bone loss. Advanced frameworks incorporating self-supervised learning and precise semantic segmentation (e.g., U-Net, TransUNet) facilitate objective measurements, such as radiographic bone loss ratios. Nevertheless, significant hurdles remain, including class imbalance, non-standardized dataset annotations, single-center study biases, and the uninterpretable "black box" nature of complex models. Addressing these challenges through multi-center validation datasets, standardized annotation protocols, and chairside integration into clinical workflows is essential for advancing automated dental diagnostics toward real-world deployment.

کلیدواژه ها:

Deep Learning ، Dental Caries ، Periodontal Disease ، Radiographic Imaging ، Convolutional Neural Networks ، YOLO ، Vision Transformers ، Semantic Segmentation