Fundus to fluorescein angiography video generation as a retinal generative foundation model

Fundus fluorescein angiography (FFA) is crucial for diagnosing and monitoring retinal vascular issues, but is limited by its invasive nature and restricted accessibility compared to color fundus (CF) imaging. Existing methods for converting CF images to FFA primarily focused on static image generation, often overlooking the dynamic changes in lesions that are critical for comprehensive diagnosis. In this study, we introduce Fundus2Video, an autoregressive generative adversarial network (GAN) model designed to generate dynamic FFA videos from single CF images. Fundus2Video demonstrates exceptional performance in video generation, achieving an FVD of 1497.12 and a PSNR of 11.77, with clinical experts validating the fidelity of the generated videos. Moreover, the model exhibits remarkable transferability across ten external public datasets, successfully addressing tasks including blood vessel segmentation, retinal disease diagnosis, systemic disease prediction, and multimodal retrieval, showcasing impressive zero-shot and few-shot capabilities. These findings position Fundus2Video as a robust, non-invasive alternative to traditional FFA examinations and a versatile retinal generative foundation model. By capturing both static and temporal retinal features, Fundus2Video provides transferable representations for multiple ophthalmic downstream tasks and may motivate analogous cross-modality studies in other medical imaging domains.