Revolutionary AI diagnoses eye disease and Parkinson's risk from retinal images
A new AI tool called RETFound can evaluate multiple health conditions from just a person's retinal image, according to a study published in Nature. Researchers trained RETFound using a self-supervised learning method on over 1.6 million unlabeled retinal photos to teach it what a normal retina looks like.
This foundation model can then easily learn the retinal features of specific diseases from a small number of labeled images. RETFound performed well at detecting eye diseases like diabetic retinopathy, and showed promise predicting risks for systemic diseases such as heart attacks, heart failure, stroke and Parkinson's.
Retinal images offer a window into a person's overall health since they allow viewing of the smallest blood vessels. The AI can now help diagnose conditions experts often miss due to lack of specialized knowledge. Researchers are exploring applying this technique to different medical imaging types.
While promising, the tool has limitations the authors must clearly communicate to ensure safe application. This accurate, label-efficient AI system could benefit global healthcare if properly optimized for local populations with community oversight on limitations.