Extracting metadata from fundus images for the screening of diabetic retinopathy

In this paper we introduce a metadata schema for retinal disorders based on XML technology for successful grading. Besides the standard retinal anatomic components, the schema focuses on lesions generally caused by diabetic retinopathy. Suitable metadata classes are introduced to describe both the 2D and 3D appearance of the anatomic parts and the lesions, as well. The schema also contains schematrons to be able to derive more complex diagnosis based on the amount and spatial distribution of the lesions. The flexible design of the schema makes any further completions possible to cover acquisitions considering more fundus fields or lesions indicating other diseases. We also demonstrate the applicabilty of the schema by presenting some implementation details of it in a currently developed automatic screening system.