To alleviate the conflict between daily increasing retinal disease patients and insufficient eye care services and increase the availability and accessibility of eye care services, developing artificial intelligence (AI) powered retinal fundus image reading systems has become extremely urgent. This project is proposed to solve two key questions in AI powered retinal fundus image reading system. One is how to learn a universal model for risk prediction and relevant structure segmentation towards multiple retinal diseases with existing publicly available but partially labelled datasets. The other is how to learn domain-agnostic and task-scalable representations to make the universal models continually learn new knowledges from new data without forgetting. To validate the effectiveness of the solutions, three datasets will be established and a prototype system for retinal fundus reading will be developed.

2023

2027