Yukun Zhou

Can a Natural Image-Based Foundation Model Outperform a Retina-Specific Model in Detecting Ocular and Systemic Diseases?

DINOv2, a natural image-based model, outperformed retina-specific RETFound in ocular disease detection, while RETFound excelled in systemic disease prediction. This highlights the need to match model selection with specific clinical tasks.

Independent Evaluation of RETFound Foundation Model's Performance on Optic Nerve Analysis Using Fundus Photography.

RETFound accurately predicts cup-to-disc ratio and average retinal nerve fiber layer thickness from fundus photos. This demonstrates its utility for optic nerve evaluation, even without specific training.