A review of publicly available ophthalmic image datasets, examining access, description, demographic coverage and the foundations needed for responsible AI research.
The research question
The ESCRS Digital Health Special Interest Group reviewed publicly available datasets for anterior-segment imaging, with particular attention to cataract, refractive and corneal surgery. A structured search identified 26 accessible datasets.
The review found that many datasets were small, poorly described, missing important demographic information or weighted towards healthy eyes. Those limitations matter when datasets become foundations for artificial intelligence and machine-learning systems intended for much broader populations.
Why access and description matter
Restrictive access conditions and inconsistent documentation make comparison and reuse difficult. Limited geographic and demographic representation can also reproduce bias, especially when data from regions such as South America and Oceania is scarce.
The work argues for clearer dataset design, greater international collaboration and a regularly maintained directory that makes available data easier to discover and assess. Better foundations support more robust research and, ultimately, safer diagnostic and treatment innovation.