Sept. 22, 2026
Artificial intelligence (AI) continues to rapidly reshape ophthalmology, with success ultimately hinging on factors beyond sophisticated algorithms. By strengthening bioinformatics infrastructure, Mayo Clinic is establishing the groundwork necessary to advance AI-enabled applications in research and practice.
"The key to advancing AI tools begins with creating structured, accessible clinical data," says Raymond Iezzi, Jr., M.D., a vitreoretinal surgeon at Mayo Clinic in Rochester, Minnesota. "In alignment with a broader institutional effort to develop a robust data platform capable of supporting innovation across specialties, we created the Ophthalmology Parametric Universal Search (OPUS), which is a dedicated bioinformatics system designed to organize and unlock access to clinical and imaging data."
OPUS aids in identifying patients for clinical trials, connecting eligible patients with emerging therapies, studying patient cohorts and analyzing outcomes. For example, when Mayo Clinic researchers need to train an AI system to recognize a specific disease, OPUS can identify appropriate patient populations and retrieve associated imaging datasets. These curated training sets support the development of algorithms capable of detecting disease, identifying risk factors, and potentially automating aspects of screening and diagnosis.
Dr. Iezzi points to previous Mayo Clinic research in Fuchs corneal dystrophy as an example of the power of quantitative phenotyping. Keith H. Baratz, M.D., a corneal specialist at Mayo Clinic in Rochester, Minnesota, and his colleagues combined objective measurements of disease severity with genetic data to identify the gene responsible for the condition, paving the way for new therapeutic strategies. Their findings were published in the New England Journal of Medicine in 2010. "Similar approaches could be expanded through AI-driven image analysis, creating new opportunities to connect imaging biomarkers with genomic discoveries," Dr. Iezzi says.
Another promising area is the use of "edge computing," in which machine learning algorithms operate in real time during a patient examination. Rather than analyzing images after the visit, AI systems integrated into examination equipment could continuously assess findings as images are captured.
"For example, an AI-enabled slit-lamp camera could document abnormalities throughout the eye while simultaneously quantifying their severity," Dr. Iezzi says. "During a retinal examination focused on age-related macular degeneration, a parallel algorithm might evaluate the optic nerve for glaucoma-related risk factors, helping us identify additional concerns without increasing workflow complexity."
Such tools could significantly reduce documentation burdens while standardizing how findings are recorded. "We could reduce the cognitive load on the clinician," Dr. Iezzi says. "It could provide standardization for the way we document those findings."
Standardized, quantitative documentation has implications far beyond efficiency. Consistent data collection improves the ability to identify patients with similar phenotypes, supports the development of precision medicine tools and enhances future AI training efforts.
Researchers at Mayo Clinic also are exploring how AI can improve large-scale screening. Dr. Iezzi and Lauren A. Dalvin, M.D., an ocular oncologist at Mayo Clinic in Rochester, Minnesota, along with their colleagues are evaluating machine learning algorithms for their ability to analyze retinal images and identify lesions that warrant referral to ocular oncology specialists. Published in the June 2026 issue of Translational Vision Science & Technology, their research found that they were able to create a robust diagnostic tool for melanocytic lesion detection with minimal annotation requirements.
Applied across large imaging archives, these technologies could strengthen disease surveillance and facilitate earlier intervention. "With the ability to run machine learning algorithms through every single retinal image ever taken at Mayo Clinic, we could effectively provide improvements in the way we screen for dangerous conditions," Dr. Iezzi says.
Ultimately, Dr. Iezzi sees AI as a tool that can augment the work done by clinicians. By reducing manual workloads, improving precision and uncovering patterns that may not be visible to the human eye, machine learning has the potential to enhance both patient care and scientific discovery.
The future may extend even further into the emerging field of oculomics, which uses information captured during eye examinations to provide insights into overall health. As AI becomes increasingly capable of detecting subtle ocular changes, clinicians may gain new opportunities to identify systemic disease risk factors and intervene earlier.
"Ophthalmic imaging generates vast amounts of detailed information about the eye, and machine learning tools can help transform those images into measurable indicators," Dr. Iezzi says. "We have the potential to create AI systems with the ability to convert subjective observations into objective, quantitative measurements — using the eyes as a window to the rest of the body."
For more information
Baratz KH, et al. E2-2 protein and Fuchs's corneal dystrophy. New England Journal of Medicine. 201:363:1016.
Ye RZD, et al. Choroidal melanocytic lesion detection using patch vectors with a foundational vision transformer. Translational Vision Science & Technology. 2026;15:1.
Refer a patient to Mayo Clinic.