This presentation introduces RETFound, a foundation model trained on over 1.6 million unlabelled retinal images using self-supervised learning. RETFound enables generalizable detection of both ocular and systemic diseases—such as diabetic retinopathy, glaucoma, myocardial infarction, and Parkinson’s disease—with significantly less labeled data. RETFound consistently outperforms supervised baselines and achieves high performance and label efficiency, offering scalable potential for real-world clinical AI deployment.