A team of researchers in China has developed DeepRETStroke, an innovative artificial intelligence system that analyzes retinal images to predict the five-year risk of a stroke (cerebrovascular accident, or CVA). Using a non-invasive photograph of the eye's fundus, the method employs deep neural networks and has shown promising results in multicenter studies.

During its development phase, scientists trained the model using nearly 900,000 images from the Shanghai Integrated Diabetes Prevention and Care System and the China National Diabetic Complications Study (Nature.com). The system was then validated with over 213,000 retinal images from patients in China, Singapore, Malaysia, the United States, the United Kingdom, and Denmark. The findings were striking: an AUC of 0.901 for predicting a first-time stroke, and 0.769 for recurrence risk.

This approach is grounded in the fact that retinal vasculature closely mirrors cerebral vasculature, sharing both embryonic origin and physiological features. Through analysis of microvascular features—such as arteriolar narrowing, tortuosity, and venous changes—the AI can detect early signs of vascular damage that conventional methods might miss.

Beyond DeepRETStroke, other research has supported this trend. For instance, a UK Biobank study published in the journal Heart analyzed over 45,000 cases and found that changes in retinal vessel shape, caliber, and branching correlated with higher stroke risk—achieving results comparable to traditional metrics like blood pressure and cholesterol levels. Similarly, traces of silent cerebral infarctions were detected by AI in retrospective studies, with AUC values exceeding 0.84.

A major strength of the Chinese model is its multicenter versatility, as it was validated across multiple ethnic groups, enhancing its potential for global application. In pilot prospective studies, it outperformed traditional clinical models and showed promise for guiding preventive strategies in primary care.

However, challenges remain. Standardized image acquisition protocols, integration with electronic health records, and acceptance by patients and clinicians are necessary. Further evaluation is needed to determine its effectiveness across specific population groups and diverse clinical settings before widespread implementation.

What’s next? AI-based retinal screening emerges as a non-invasive, rapid, and cost-effective tool for assessing cerebrovascular risk. Its integration into routine eye exams could revolutionize preventive medicine—allowing strokes to be anticipated before the first event occurs and potentially saving thousands of lives. The challenge now lies in validating and regulating its clinical use globally.