A Scoping Review on Synthetic Datasets: Opportunities, Challenges and Future Directions for Endodontics

Area of Science:

• Dental AI
• Medical Imaging
• Synthetic Data Generation

Background:

• Artificial Intelligence (AI) and machine learning are advancing dental diagnostics.
• High-quality, annotated datasets are scarce, posing challenges for AI development.
• Synthetic data offers a potential solution to data scarcity and privacy concerns in endodontics.

Purpose of the Study:

• To review synthetic datasets in dentistry and endodontics.
• To identify common data generation techniques and evaluate their effectiveness.
• To explore limitations and future research directions for synthetic data in endodontics.

Main Methods:

• A systematic search of major databases (PubMed, Scopus, EBSCO, Google Scholar) was performed.
• Data on image modalities, generation models, dataset sizes, and evaluation metrics were extracted.
• Descriptive analysis was used to summarize findings on synthetic data in endodontics.

Main Results:

• Eleven studies were included, predominantly using Generative Adversarial Networks (GANs).
• Panoramic radiographs were the most common modality; synthetic data alone showed moderate fidelity.
• Combining synthetic with real data improved AI model accuracy for tasks like canal classification and implant segmentation.

Conclusions:

• Synthetic data is an emerging tool in endodontics with significant potential.
• Future research should focus on diverse, clinically relevant datasets using advanced models like diffusion and hybrid approaches.
• Synthetic data can enhance AI-driven diagnostics, education, and endodontic care.