A new artificial intelligence system can analyze detailed 3D images of the eye's retina and spot signs of disease faster and more accurately than human review alone, potentially catching vision-threatening conditions before they cause permanent damage. Researchers at Washington University School of Medicine, the University of Washington, and Genentech developed OCTCube-M, a family of three AI models trained on over 26,000 three-dimensional eye scans to identify eight different retinal diseases, including age-related macular degeneration, the leading cause of blindness in people over 50.

Why Are Eye Scans Producing So Much Data That Doctors Can't Keep Up?

Modern eye imaging technology called optical coherence tomography (OCT) has revolutionized how doctors diagnose vision problems. A single scan produces hundreds of cross-sectional images that create a detailed 3D picture of the retina and optic nerve, revealing early signs of diseases like glaucoma, macular degeneration, and diabetic retinopathy. But that precision comes with a cost: physicians must manually review hundreds of images per scan, a time-consuming process vulnerable to human error.

According to the World Health Organization, at least 2.2 billion people worldwide have vision impairment, making the need for faster, more accurate diagnosis urgent. The volume of imaging data has outpaced doctors' ability to process it efficiently, creating a gap where subtle signs of disease can slip through.

How Does OCTCube-M Improve on Older AI Models?

The new system represents a significant leap forward in AI-assisted eye care. When compared to models trained only on 2D retinal images, OCTCube-M more accurately identified six of the eight retinal diseases by approximately four to six percentage points. In practical terms, that translates to the tool finding 43 to 60 additional cases of eye disease out of every 1,000 individuals screened.

The researchers trained OCTCube-M on 1.62 million individual retinal slices from over 26,000 three-dimensional scans. They then enhanced the model by adding data from two other eye imaging techniques: infrared retinal imaging and fundus autofluorescence imaging. By combining all three imaging types, the AI system could construct a more complete picture of what's happening inside the eye.

The results were particularly striking for geographic atrophy, a severe form of macular degeneration affecting about 5 million people worldwide. The model trained on all three imaging types outperformed the current state-of-the-art model by an average of nearly 50% when predicting how fast the condition would progress.

What Are the Real-World Benefits for Patients and Doctors?

The implications extend far beyond faster diagnosis. By accurately predicting disease progression, researchers can design smaller, more efficient clinical trials for new treatments, potentially lowering costs and shortening the time it takes to test new therapies. This means fewer people would be exposed to treatments that don't work, and effective drugs could reach patients sooner.

The AI system also demonstrated an unexpected capability: it could infer health risks beyond the eye itself. The model predicted outcomes such as heart attack, stroke, and kidney failure based solely on retinal imaging. This is possible because the tiny blood vessels in the retina are anatomically similar to those in the kidney, and the processes that lead to plaque buildup in blood vessels feeding the heart and brain leave visible signatures in the eye.

"Today's eye scans provide physicians an unprecedented, highly detailed view of the inside of the eye, revealing structures and subtle changes that would otherwise go undetected. But we still lack the tools to help physicians process the volume of generated images. Our AI system has the potential to empower physicians to make faster diagnoses, tailor treatment more precisely and design clinical trials that bring new therapies to patients faster," said Aaron Lee, MD, the Arthur W. Stickle Distinguished Professor of Ophthalmology and Visual Sciences at Washington University School of Medicine.

Aaron Lee, MD, Arthur W. Stickle Distinguished Professor of Ophthalmology and Visual Sciences at Washington University School of Medicine

Steps to Responsible AI Adoption in Healthcare

While OCTCube-M shows tremendous promise, its development reflects a broader shift in how healthcare institutions are approaching AI. At UC Davis Health, leaders have established a rigorous framework for ensuring AI tools are safe, fair, and effective before they reach patients.

• Rigorous Pre-Adoption Review: An interdisciplinary team of healthcare providers and data scientists analyzes potential AI applications and their use prior to adoption to evaluate whether they are safe enough for patients. After a trial period, the group reviews performance to ensure outcomes match expectations.
• Equity-Focused Development: UC Davis Health developed BE-FAIR, a Bias-reduction and Equity Framework for Assessing, Implementing, and Redesigning healthcare predictive models. This framework promotes anti-racism, community engagement, and inclusion of historical context when AI is developed and implemented.
• Data Security and Privacy Protection: A dedicated data security team reviews and pressure tests AI applications to ensure patient privacy is protected and patient data remains secure. A multidisciplinary committee of clinicians, scientists, privacy experts, and legal specialists sets policies to prevent unauthorized data sharing.

The philosophy underlying these safeguards is clear: AI should assist doctors and nurses, not replace them. At UC Davis Health, AI makes recommendations, but physicians make the final call on medical decisions. This human-centered approach reflects a growing recognition that the most effective AI in healthcare is one that enhances what doctors can do, rather than attempting to substitute for their judgment.

What Comes Next for AI in Eye Care?

The researchers at Washington University plan to train OCTCube-M with larger datasets encompassing more patients, more diseases, and additional types of imaging data to continue improving the system. As the model learns from more diverse populations and clinical scenarios, its accuracy and reliability should increase further.

The potential impact is substantial. By turning a routine eye exam into a tool for detecting systemic diseases like heart disease and kidney failure, AI could enable earlier intervention and better outcomes for millions of patients who might otherwise go undiagnosed until their conditions are far more advanced. For the 2.2 billion people worldwide with vision impairment, faster and more accurate diagnosis could mean the difference between preserving sight and permanent vision loss.