AI detects early eye changes signaling diabetic retinopathy before clinical symptoms

Diabetic retinopathy—damage to the blood vessels in the retina caused by diabetes—is the most common eye disease among people with diabetes and a leading cause of blindness in American adults. The numbers are sobering: between 2010 and 2050, the number of Americans living with the condition is expected to nearly double, from 7.7 million to 14.6 million. Yet for years, the disease has had a crucial vulnerability: by the time a standard clinical eye exam detects it, significant damage has already occurred. Researchers at Indiana University have now developed a method to change that timeline.

The key insight is that diabetes begins harming the eye long before any visible symptoms appear—changes that traditional examination cannot yet see. Ann E. Elsner, a professor at the Indiana University School of Optometry, and her team discovered that these early alterations can be measured using specialized optical imaging techniques paired with artificial intelligence analysis. The work, published in PLOS ONE and funded by the National Institutes of Health's National Eye Institute, suggests a path toward catching the disease in its infancy, before irreversible vision loss takes hold.

Elsner and her colleagues, including PhD student Joel A. Papay, analyzed retinal images collected from volunteers with diabetes and healthy control subjects, supplemented by screening data from underserved communities at UC Berkeley and Alameda Health. The computer algorithms they developed work differently from many existing AI approaches to diabetic retinopathy detection. While other systems cast a wide net—identifying any image features that differ between diabetic and non-diabetic eyes—Elsner's method targets specific retinal layers and tissue types. This precision matters because it allows the algorithm to surface information that other systems overlook entirely, information that has been sitting in standard clinical images all along but simply ignored.

"Early detection of retinal damage from diabetes is possible to obtain with painless methods and might help identify undiagnosed patients early enough to diminish the consequences of uncontrolled diabetes," Elsner said. The implications ripple outward. If biomarkers can be detected before symptoms emerge, physicians gain the ability to intervene earlier, potentially preventing or slowing vision loss. The method could also identify people with undiagnosed diabetes—catching the disease through eye changes before a patient even knows they have it.

What makes this advance particularly significant is that it does not require new equipment or new imaging techniques. The retinal images used in the study are the kind already collected in well-equipped clinics across the country. The innovation lies in how the data is processed and what the algorithm is trained to see. By combining this new method with other AI approaches, clinicians could layer early-stage biomarker information onto existing detection systems, creating a more complete picture of retinal health at earlier stages of disease.

The expected doubling of diabetic retinopathy cases over the next three decades makes this work urgent. Early detection offers a concrete tool for slowing or preventing blindness in millions of Americans, particularly in underserved populations where access to eye care is already limited. The next phase will likely involve testing the method more broadly and integrating it into clinical workflows—turning a laboratory finding into a tool that actually reaches patients before their vision begins to fade.