Advancements in Caries Diagnostics Using Bitewing Radiography: A Systematic Review of Deep Learning Approaches.

Area of Science:

• Dentistry
• Radiology
• Artificial Intelligence

Background:

• Deep learning (DL) techniques are increasingly utilized for improving radiographic caries diagnosis.
• Bitewing radiographs are a key modality for detecting dental caries.

Purpose of the Study:

• To systematically review the application of deep learning for caries diagnosis using bitewing radiographs.
• To assess the diagnostic performance, methodologies, and biases of DL models in caries detection, segmentation, and classification.

Main Methods:

• A systematic review following PRISMA guidelines was conducted.
• Literature searches were performed on Web of Science and PubMed for studies predating March 2025.
• Data extraction focused on model architectures, datasets, performance metrics, and bias using QUADAS-2.

Main Results:

• Twenty-three studies were included, evaluating DL for caries detection, segmentation, and classification.
• Common DL models included ResNet and YOLO; dataset sizes varied significantly.
• High diagnostic accuracies (70-99%) were reported, with some AI models outperforming human experts.
• Significant variability in methodologies, performance metrics, and reporting standards was noted.
• All studies exhibited a high risk of bias in at least one domain.

Conclusions:

• Deep learning models demonstrate potential as assistive tools for caries diagnosis on bitewing radiographs.
• Challenges include methodological heterogeneity, lack of standardization, limited dataset diversity, and bias.
• Further research is needed to address these limitations for robust clinical implementation.