Ophthalmology has been a proving ground for emerging machine-learning
paradigms in medicine. The efforts of the ophthalmic and computer science communities have yielded thousands of
innovative models, exploring tasks ranging from basic science to clinical epidemiology.
The Open Ophthalmology Database was developed in 2023 as the Vision Intelligence Network to provide a systematic
means of identifying and
implementing machine-learning models in ophthalmology. It leverages natural language processing to survey the
published literature for pertinent models and characterize them in terms of their purpose, data substrate, and
algorithmic approaches.
Key data cataloged in the database include the following:
- Model: the name of the model, defaulting to the primary author if model name is undetermined
- Subspecialty: the ophthalmic subspecialty to which the model is most relevant
- Title: the title of the corresponding publication
- Task: the primary function of the model
- Task type: the class of objective toward which the model is trained
- Algorithm type: the primary class of algorithm used in the model
- Data type: the primary data type used in training
- Dataset size: the number of training examples
- Paper link: link to corresponding publication
- Code: link to provided codebase
- Date: publication date
- Code identified: whether or not a corresponding codebase was identified
Curation relies on hard-coded metadata extraction where possible. Where this is not feasible, the model invokes agentic tools to summarize pertinent data. While we make efforts to check the reliability of these outputs, we note that such tools are fundamentally probabilistic and may generate inaccurate summaries. All entries are linked to their respective publications, and users are advised to refer to the reference for all technical details regarding the model. This database is an experimental research effort and should not be used for clinical decision-making.
Inquiries are welcome at contact@openophthodb.org.