Researchers' New AI Spots Eye Disease Early From Retina Photos

A team of researchers has developed an AI system that can look at a photo of your retina—the light-sensitive tissue at the back of your eye—and detect early signs of diseases that can lead to blindness. These include diabetic retinopathy, a complication of diabetes; glaucoma; and age-related macular degeneration. Normally, trained eye specialists have to study these images closely, which takes time and experience. The new AI does it automatically and, in testing, was right about 94% of the time.

The researchers did something else important: they made the AI "explain itself." Many AI systems are "black boxes"—they give a yes-or-no answer but stay silent on how they reached it. This system produces heatmaps that highlight the exact areas of the eye it used to make its call. If an eye doctor disagrees, they can see what the AI was looking at and judge whether it makes sense. This kind of transparency builds trust and helps catch mistakes.

In a second part of the project, the same AI was taught to trace the tiny blood vessels in the eye. That 'map' matters because conditions like diabetic retinopathy damage those vessels. The researchers tested dozens of different AI architectures, including newer "attention" models that learn which parts of an image matter most. Their best vessel-tracing tool outperformed older methods on both public datasets they used, FIVES and DRIVE.

The catch: this is still research. The system works well on stored images from a lab, but it hasn't yet been tested in busy clinics on real patients with varying skin tones, camera brands, and image quality. Still, if these tools survive real-world testing, they could be a game changer—especially in rural areas and developing countries, where eye specialists are scarce. Patients could get a quick photo screening, and an AI could flag those who need a human expert right away. That simple process could save millions of people from preventable blindness.