Automatic retinal vessel extraction from fundus images taken from patients with diabetes

We present here a novel database of retinal fundus images for the automatic extraction of retinal surface vessels. In contrast to other publicly available databases such as the DRIVE [Staal et al, 2004, IEEE Trans Med Imaging, 23, 501-509] or STARE database [Hoover et al, 2000, IEEE Trans Med Imaging, 19, 203-210], our database consists of only fundus images recorded from patients with diabetes. As the incidence of diabetes is increasing worldwide [Zimmet et al, 2001, Nature, 414, 782-787], screening patients for diabetic retinopathy is becoming a more onerous task. In addition to diabetes, these patients showed various pathologies such as age-related macular degeneration, hypertension and/or glaucoma. Therefore, our database is divided along the lines of retinal pathology. All these images were obtained from a diabetic retinopathy screening programme. Either a Canon CR DGi, a Topcon NW6S, or a Topcon NW8 fundus camera were used to capture the retinal images with a field of view of 45 degrees. Two separate unsupervised vessel extraction methods [Holm and McLoughlin, 2012, Perception, 41 ECVP Supplement, 103] were applied to this novel database and compared to the manually segmented images.