Explainable Artificial Intelligence for Automated Diagnosis and Risk Assessment in Dental Caries and Periodontal Diseases
Main Article Content
Abstract
Machine learning and deep learning-based artificial intelligence (AI) technologies are being considered as additional tools for dental diagnostics. Caries and periodontal diseases are examples of cases where diagnosis involves the combination of clinical parameters, radiographic and intraoral images, and longitudinal patient data. At the same time, high-precision AI models can be considered black boxes, thus limiting the understanding of their predictions and relevance to clinical practice. Explainable artificial intelligence (XAI) can make the predictions more transparent and understandable. The aim of this narrative review is to explore the use of XAI for automatic dental caries and periodontal disease diagnosis and risk assessment. This review examines AI architecture types, dental imaging modalities, clinical and non-image-based data, and main XAI approaches used to explain the predictions, such as SHAP, LIME, Grad-CAM, saliency maps, and attention. Based on the existing research, promising diagnostic performance of AI algorithms in caries detection based on intraoral and radiographic images and the potential of AI-based periodontal disease detection and classification are seen. However, the available literature is limited by retrospective design, heterogeneity of the reference standard, lack of external validation, dataset shift, lack of reporting, and poor evaluation of explanation quality. A 2026 systematic review included 19 articles, with 18 high-risk-of-bias articles and only one article with an assessment of human understanding of explanations. Therefore, the clinical evaluation of AI requires the consideration of not only the diagnostic performance but also calibration, external validation, explanation quality and reliability, fairness, uncertainty, and human-centered evaluation.
