We have experienced some major problems with our database server and can only provide you with the image files by download. View data, data annotation, registration, and submission are no longer possible. Datasets may be downloaded using the links below.)
The ROC aims to help patients with diabetes through improving computer aided detection and diagnosis (CAD) of diabetic retinopathy. Diabetic retinopathy is the second largest cause of blindness in the US and Europe. Most visual loss and blindness from diabetic retinopathy can be prevented through early diagnosis or screening.
Computer algorithms have been developed in order to detect the signs of diabetic retinopathy from retinal images obtained using a digital retinal camera. However, few, if any of these algorithms have been applied in clinical practice. ROC facilitates the translation of diabetic retinopathy CAD into clinical practice by:
Currently, we have released a first data set, aimed at CAD of microaneurysms and dot hemorrhages. These abnormalities are amongst the first signs of the presence of diabetic retinopathy. For some example lesions see Figure 1 below. On this site, interested research groups and companies can register a team, download the data and submit results (multiple submissions possible). The results consist of a set of image-locations combined with a 'degree of suspicion' for each location that it is an abnormality. Each of the submitted results will be analyzed in the same way using FROC analysis. Based on this each submission will be awarded a score.
Several teams have submitted their system's results to the competition website, a number of them were presented at the CAD Conference of SPIE: Medical Imaging 2009. The results were presented as regular conference contributions and in a special workshop included in that meeting. Since then, more teams have submitted results and the competition website remains active and open for new submissions. An overview paper describing the results of the competition, jointly authored by the first set of teams was published in IEEE Transactions on Medical Imaging. The ROC micoaneurysm dataset will remain available online and more results can be submitted via this website.
Figure 1: Details from retinal images with lesion markers.
The content of this website is copyright 2007 (c) The University of Iowa and the ROC organizers (Michael D. Abramoff). All rights are reserved. Permission is granted to others to use the ROC image data for developing and testing of diabetic retinopathy algorithms purposes only as described in the competition rules. Any use of the datasets, although free, must reference the ROC (see competition rules). The datasets may not be changed in any way, nor may derivate works be produced without express written permission. Submitted results may be used for further research. No commercial permission is granted without a separate written license from the University of Iowa. Funded in part by the National Eye Institute (5R01EY017066), Research to Prevent Blindness, NY, NY and the Netherlands Organization for Health Related Research.