Brief

Researchers have developed EFIQA, a novel framework for assessing the quality of fundus images that does not require quality-related supervision. Instead of learning from human-annotated labels, EFIQA utilizes anatomical priors to identify regions of degradation. The framework employs an unsupervised anomaly detector trained via masked anatomical inpainting to pinpoint missing vasculature, then distills this knowledge into an adapter that maps features from a frozen foundation model to precise quality maps. Evaluations on external datasets show that this label-free approach achieves superior performance and explainability compared to supervised methods, making it suitable for real-world applications. AI