New AI Method Uses Eye Scans to Predict Alzheimer's Risk

A new research paper on arXiv introduces REVEAL++, an advancement in vision-language AI models that uses retinal fundus images—standard eye scans—combined with structured clinical risk narratives to predict Alzheimer's disease risk. The key innovation is "differentiable phenotypic grouping," a method that groups individuals with similar risk profiles as multi-positive pairs during contrastive learning, improving the model's ability to detect subtle structural patterns in the eye associated with future cognitive decline.

This matters because the retina offers a noninvasive window into neurodegenerative disease. By pairing an eye scan with a patient's clinical data, the model can identify early warning signs of Alzheimer's before symptoms become severe. Early detection could enable earlier intervention and potentially slow disease progression, improving quality of life.

The model builds on earlier work called REVEAL, and the researchers focused on refining how the AI handles patients with similar but not identical risk factors. Instead of treating all patients as independent examples, the new grouping strategy helps the model learn more robust representations of disease risk.

If you're curious, you can read the full research paper on arXiv. While the tool is not yet available for clinical use, it represents a promising step toward non-invasive, accessible early screening for Alzheimer's disease.