Journal of Dental Medicine

Background and Aims: Dental caries is one of the most prevalent chronic oral diseases worldwide. Timely and accurate diagnosis of dental caries plays a crucial role in preventing lesion progression and reducing complications. This study aimed to systematically review the studies on dental caries detection using machine learning algorithms applied to periapical radiographs.
Materials and Methods: A comprehensive search was conducted in PubMed, Scopus, Web of Science, IEEE Xplore, and Google Scholar databases up to the end of 2024. Inclusion criteria comprised studies using machine learning algorithms for detecting dental caries in periapical or intraoral radiographs. The quality of studies was assessed using the QUADAS-2 tool.
Results: From 825 initial articles, 13 studies met the inclusion criteria. All studies used Convolutional Neural Networks (CNNs) with various architectures including ResNet, VGG, Inception, DenseNet, and YOLO. ResNet-based models and their hybrid variants showed the best performance with diagnostic accuracy ranging from 82% to 98%. Comparison with human experts in 6 studies revealed that deep learning algorithms demonstrated similar or superior performance.
Conclusion: From the results, deep learning especially convolutional neural networks, had significant potential for improving dental caries detection in periapical radiographs. However, challenges such as limited high-quality training data and generalizability issues need further investigation.