DR-MERGE: A Large-Scale Multi-Source Retinal Image Dataset for Diabetic Retinopathy via Systematic Integration and Deduplication

Diabetes can hurt eyes badly, causing what we call diabetic retinopathy. This eye problem is a big reason people lose sight around the world. To make computer programs that help doctors find this eye problem, we need lots of eye pictures. Right now, available picture groups are often too small or not varied enough. This makes building good detection systems tough. Our work introduces DR-MERGE, a new large group of eye pictures we made by joining four public sets: Diabetic Retinopathy- Glaucoma-Mask, Diabetic Retinopathy Dataset, ODIR-5K, and STARE project. We made a careful way to put these together, fixing issues like different file types, finding same pictures using MD5 codes, and organizing everything well. Our method found and joined 17,976 different eye pictures from starting 18,104 pictures, taking out 128 copies that were exactly the same. DR- MERGE is now one of the biggest joined sets of eye pictures made just for diabetes eye problem research. We explain our full method, show numbers about the original sets and joined collection, and talk about how this can help move forward research in finding eye problems using computers. The picture set and our code are out there for others to check our work and keep studying medical pictures.