Machine Learning and Artificial Intelligence Models Applied to the Detection and Classification of Dental Caries: A Systematic Literature Review

Abstract

Objective: To systematically evaluate the performance of artificial intelligence (AI) and machine learning (ML) models for the detection and classification of dental caries across different imaging modalities. Review methodology: A systematic literature search was conducted in PubMed/MEDLINE, Scopus, Web of Science, IEEE Xplore, and Google Scholar covering publications from January 2015 through March 2026. Studies reporting quantitative performance metrics for AI/ML models applied to caries detection, classification, or segmentation were included. A three-stage screening process (title, abstract, full-text) was applied following PRISMA guidelines. Results: A total of 45 studies were included. Deep learning, particularly convolutional neural networks, constituted the predominant approach (~69%). The most employed architectures include the YOLO family (v3–v11), U-Net variants, and transfer-learning classifiers (ResNet, VGG, DenseNet, EfficientNet, MobileNet). Reported accuracy ranged from 73.3% to 98.6%, with pooled meta-analytic sensitivity of 0.85 and specificity of 0.90. Only 12% of studies used public datasets. Conclusions: AI/ML systems demonstrate substantial diagnostic potential for dental caries, frequently matching clinician performance. Critical gaps persist regarding multicenter validation, clinically interpretable models, and prospective trials evaluating patient outcomes.