Publicly Available Dental Image Datasets for Artificial Intelligence.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging
• Data Science

Background:

• Artificial intelligence (AI) in dentistry relies heavily on large, well-annotated datasets.
• The availability and characteristics of public dental imaging datasets are not well-documented.
• This lack of accessible data hinders the development and validation of AI tools for dental applications.

Purpose of the Study:

• To conduct a comprehensive overview of publicly available dental imaging datasets.
• To identify gaps and challenges in existing datasets to support AI development in dentistry.
• To assess the FAIRness (Findability, Accessibility, Interoperability, Reusability) of these datasets.

Main Methods:

• An observational study involving a comprehensive search of public dataset resources, academic databases, preprints, and AI challenges.
• Focus on dental AI datasets containing images (e.g., radiographs, intraoral scans) published between 2020-2023, with PubMed searches extending to 2011.
• Extraction of dataset characteristics including size, imaging modality, annotations, ethical clearance, licensing, and FAIRness metrics.

Main Results:

• Only 16 unique dental imaging datasets were identified from over 131,000 records.
• The primary focus of datasets was tooth segmentation and labeling, with panoramic radiography being the most common modality.
• Significant inconsistencies were found in metadata reporting, ethical approval (31.2%), and licensing (43.75%), impacting data usability and FAIRness.

Conclusions:

• There is a notable scarcity of high-quality, publicly available dental imaging datasets.
• Inconsistent metadata, lack of ethical considerations, and poor FAIRness metrics limit the potential of current datasets for AI development.
• Future efforts must prioritize increasing data availability, diversity, metadata completeness, and adherence to FAIR principles to advance AI in dental diagnostics and research.