A growing class of medical AI systems can flag signs of disease from a photograph alone, with no blood draw or laboratory test. The clearest example in routine clinical use is EyeArt, a system from the US company Eyenuk that examines a photograph of the retina and reports whether a patient shows signs of diabetic retinopathy, a complication of diabetes that can lead to blindness if it is not caught in time.
The approach is sometimes grouped under the label “visual diagnosis”: training a model on large sets of medical images, labeled by clinicians, until it can recognize the visual patterns associated with a condition. Eye photographs have proved a particularly productive target, because the retina is one of the few places where blood vessels and nerve tissue can be photographed directly and because the damage caused by diabetes follows well-documented visual patterns.
What EyeArt does
According to Eyenuk, EyeArt takes high-resolution images of the back of the eye, captured with a standard retinal camera, and returns a result within seconds indicating whether the patient has more than mild diabetic retinopathy and should be referred to an eye specialist. The company describes the system as autonomous, meaning the screening result is produced by the software without a clinician interpreting the image.
The US Food and Drug Administration cleared EyeArt for autonomous detection of diabetic retinopathy in 2020, and Eyenuk has since obtained additional clearances expanding the list of compatible cameras. FDA clearance means the device can be marketed for use in clinics and primary care settings in the United States under the conditions set out in the clearance.
EyeArt is not the only system of its kind. Other companies have also received FDA authorization for autonomous diabetic retinopathy screening, and the category is now established enough that professional guidelines in the United States discuss AI-based screening as an option.
Why it matters for access to care
The practical case for these systems rests on a shortage of specialists. People with diabetes are advised to have their eyes examined regularly, but many never see an ophthalmologist, either because there is none nearby or because the appointment is easy to postpone. A screening tool that can run in a primary care clinic or a pharmacy, operated by staff who are not eye specialists, lowers that barrier. Patients who screen positive are referred onward; those who screen negative are told to return for their next routine check.
In regions with few ophthalmologists, that model could extend screening to populations that currently go unscreened, which is the argument Eyenuk and its peers make for the technology.
Limits and open questions
Autonomous screening does not diagnose or treat disease; it identifies who needs to see a doctor. Performance depends on image quality, on the camera used, and on whether the patient population resembles the data the model was trained on. Regulators have so far cleared these systems for narrowly defined tasks, and claims that AI can read heart disease or other conditions from a face photograph remain research findings rather than approved clinical tools.
Even within those limits, retinal screening shows what image-based medical AI looks like when it reaches patients: a narrow, validated task, performed quickly at the point of care, with a human specialist still responsible for anything the system flags.
