The Eye Can Reveal Your Health: Russian Researchers Develop AI to Analyze the Fundus
Russian scientists have developed a system that analyzes fundus images and assesses the probability of 15 diseases and conditions, ranging from glaucoma to hypertension and AIDS. The technology is designed to significantly improve the quality of diagnostics.
The eyes are more than a mirror of the soul – they can also provide valuable information about a person’s health. Simple images of the retina and iris can reveal changes associated with vascular health and the function of different systems throughout the body. That principle underpins a new technology developed by researchers at Skoltech, the Z-union AI consortium and Sber AI Lab.
The system developed by the Russian researchers uses computer vision. From a fundus photograph, the algorithm assesses the probability that a patient has any of 15 diseases and pathological conditions. The list includes ophthalmic diseases such as glaucoma, cataracts and diabetic retinopathy. The AI also looks for changes associated with hypertension, atherosclerosis, lupus and AIDS.
What Does the Neural Network See?
To train the model, the researchers collected, annotated and de-identified more than 20,000 fundus images from different people. The dataset included 12,000 images from open sources and about 8,350 clinical images collected by the researchers themselves.
To make the task more challenging, the researchers deliberately added rare pathological cases. This is an important consideration when training medical AI. An ordinary image of a healthy eye does little to teach a system how to recognize a rare disease. The algorithm needs to see enough examples to identify patterns that may not always be apparent to the human eye.
As a result, the model learned to assess the probability of each of the 15 conditions from a single image. On the data used in the study, performance measured by ROC AUC reached 0.997, with 1.0 representing the highest possible score. The technology still needs to undergo full clinical validation in larger patient populations.
From Vision to Overall Health
Perhaps the most significant aspect of the project is its potential to change the approach to medical diagnostics by making it more accurate, faster and more effective. That could benefit both physicians and patients.
A person might, for example, visit an ophthalmologist because of worsening vision. A fundus image taken by the physician could reveal an early-stage condition the patient did not even know was developing because the retina can reflect changes occurring elsewhere in the body. As a result, the patient could receive not only recommendations for further treatment of the problem that prompted the visit but also a referral for a more detailed medical evaluation.
According to study co-author Yulia Sarana, a researcher at Skoltech’s Center for Bio- and Medical Technologies, the researchers hope that combining this noninvasive approach with existing diagnostic methods will make it possible to detect serious diseases at earlier stages, begin treatment promptly and ultimately improve quality of life.
Medical AI in Practice
More broadly, computer-based analysis of fundus images is emerging as an active area of development in Russia. In 2023, Sechenov University introduced RetinAIcheck, a system designed to analyze fundus images and detect hypertensive retinopathy. In June 2026, the platform received a registration certificate from Roszdravnadzor, Russia’s healthcare regulator.
Technologies of this kind could be particularly useful when large numbers of people need preliminary screening. AI can rapidly review images and sort them by risk level, leaving specialists to make the final determination.
Testing in Clinical Practice
The next step is clinical validation in large patient populations. Researchers need to evaluate the algorithm in large and diverse groups of people, including under routine outpatient clinical conditions.
If systems like this successfully undergo clinical validation and are adopted in medical practice, computer vision could gradually move from an experimental technology to a routine tool for physicians. That would allow AI to help not only after a disease has fully manifested itself but at a stage when patients still have more opportunities to begin treatment and take steps to protect their health.
