AI to Prevent Blindness: The Tool Created by Argentinians to Detect a Diabetes Complication Early
It was developed by CONICET researchers together with specialists from Hospital El Cruce, among others; it uses artificial intelligence to make an early diagnosis and prevent irreversible vision loss.
- June 30, 2026
- 16:45
- 5 minutes reading
Alejandro Horvat
LA NACION
Listen to Note
Follow on
Diabetic retinopathy is one of the leading causes of preventable blindness in working-age adults. The most concerning thing is that it usually advances silently. During the early stages, it does not cause discomfort or visual alterations, so many people arrive at the doctor's office when the damage is already irreversible.
Faced with this scenario, an Argentine development based on artificial intelligence (AI) seeks to change the logic of diagnosis and bring eye check-ups to places where there are currently no specialists.
This is Retinar, a screening platform that allows identifying patients at risk of diabetic retinopathy from a photograph of the fundus taken by a trained technician. Then, an algorithm analyzes the image in seconds, prioritizes cases that require attention, and refers them to an ophthalmologist to confirm the diagnosis via telemedicine. The tool does not replace the doctor, but functions as an intelligent filter to expand access to screening.
The project originated from the Transformar Salud program, an initiative aimed at developing technological solutions for concrete problems in the healthcare system. In that framework, CONICET researchers, specialists from Hospital de Alta Complejidad El Cruce and professionals from Hospital Julieta Lanteri de Tandil, Universidad Nacional del Centro (UNICEN) and PLADEMA/Yatiris, among other teams, developed a first prototype that began testing in Florencio Varela.
Two years later, the experience was implemented in Tandil, where daily use allowed the platform to be perfected and progress toward a scalable model for other regions of the country.
"Diabetic retinopathy is an eye complication caused by the chronic damage that high blood sugar levels produce on the small blood vessels of the retina," explains Mercedes Leguía, Head of Ophthalmology at Hospital El Cruce. According to her, initially those vessels weaken and can leak fluid or blood.
In more advanced stages, some become blocked and the eye tries to compensate for the lack of oxygen by forming new extremely fragile blood vessels, which can bleed, create scars, and even cause retinal detachment.
An Asymptomatic Start
"The damage is silent. In its initial stages, it presents no symptoms or pain. When the patient waits to notice vision loss to consult, the lesions are usually severe and irreversible," warns the specialist. Therefore, periodic fundus examination is one of the main strategies to prevent blindness associated with diabetes.
However, accessing that study is not always easy. In many regions of the country, ophthalmologists are concentrated in large urban centers and getting an appointment can take months. Added to this are the logistical difficulties: traveling, losing a workday, and often going accompanied, since the traditional exam requires dilating the pupil and then vision remains blurry for several hours. According to data cited by the team, six out of 10 people with diabetes fail to complete the annual eye check-up recommended by the World Health Organization.
Retinar tries to solve precisely that gap. The procedure begins at a health center, where a nurse or trained technician obtains an image of the fundus using a non-mydriatic retinograph, a device that does not require dilating the pupil. The photograph is uploaded to the platform and the algorithm performs two consecutive analyses: first it verifies that the image is of sufficient quality to be interpreted and then determines if there are lesions compatible with diabetic retinopathy.
The system automatically classifies each study as "Non-referable" when the eye is healthy or has mild lesions, or "Referable" when it detects signs of moderate or advanced disease requiring specialist evaluation. Additionally, it generates a heat map that indicates the areas of the image that prompted the classification, highlighting lesions such as microhemorrhages, exudates, or cotton-wool spots.
"The core of Retinar is an artificial intelligence algorithm trained using neural networks that classifies images into two very clear categories: referable and non-referable," explains Leguía. She adds that this classification "allows us to quickly prioritize those who need to be evaluated by a specialist."
The doctor emphasizes that the tool does not replace professional judgment. "Retinar's role is to assist and prioritize referable patients. The final clinical decision always depends on the ophthalmologist, who confirms the diagnosis, determines the severity of the disease, and defines the treatment," she notes.
No False Negatives
One of the most relevant results of the project was the validation obtained during territorial implementation. When comparing the classifications made by artificial intelligence with the evaluation of the reporting ophthalmologist, the system achieved 100% sensitivity, that is, no false negatives were recorded among the evaluated cases. In other words, it did not fail to identify patients who required referral.
For researchers, this capability opens the possibility of reorganizing ophthalmic care. Instead of relying exclusively on in-person consultations with specialists, primary care centers could perform screening locally and refer only those who truly need more complex studies or treatments. That way, the ophthalmologist's time is concentrated on higher-risk cases.
The initiative continues to evolve. Since November, the team incorporated 19 new features, launched five updated versions of the platform, presented results at scientific conferences, submitted a paper for an international award, and is advancing in the approval process before ANMAT. Simultaneously, it is developing an implementation manual to facilitate other provinces in replicating the model.
By Alejandro Horvat
- Health
- Artificial intelligence
