Abstract
Presentation Description :
Ophthalmology is among the most technology-driven of the all the medical specialties, with treatments utilizing high-spec medical lasers and advanced microsurgical techniques, and diagnostics involving ultra-high resolution imaging. Ophthalmology is also at the forefront of many trailblazing research areas in healthcare, such as stem cell therapy, gene therapy, and - most recently - artificial intelligence. In my presentation, I will describe the development of RETFound, a foundation model for ophthalmology. RETFound was trained on 1.6 million retinal images using self-supervised learning and utilising a vision transformer architecture. In a recent publication in Nature, we demonstrate that RETFound performs better that other approaches for a diverse range of downstream clinical tasks, from diabetic retinopathy screening to predicting progression of age-related macular degeneration, to using the eye as a window to systemic disease (“oculomics”). We also show that RETFound is more robust on external validation, fairer across ethnicities, and more label efficient, opening the possibility of its use in less common retinal disease. We will describe the process to create this foundation model as well as our plans to scale and validate the model, going from nearly 2 million to 20 million images, and making it both 3D and truly multi-modal.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.