A General Model for Retinal Segmentation and Quantification
An AI trained on 200,000 eye images spots over 20 different conditions from a single scan.
Deep Dive
Researchers have developed RetSAM, an AI framework that analyzes standard retinal scans to detect and quantify signs of disease. Trained on over 200,000 images, it segments anatomical structures and identifies more than 20 distinct lesion types, converting them into over 30 standardized health biomarkers. It outperforms previous methods by nearly 4% on average and excels at multi-disease analysis, including for diabetic retinopathy, glaucoma, and macular degeneration.
Why It Matters
This could enable large-scale, affordable screening for multiple eye and systemic diseases from a single, common test.