Automatic detection of red lesions in digital color retinal images

These days, Automated screening systems are becoming the best possible solution for not only reducing burden on the medical practitioners but also reducing human prone errors with the help of accurate automated diagnosis of the disease. In Diabetic Retinopathy (DR), the crucial step for an automated system is exhaustive detection of the lesions. In this paper, a modified red lesion detection method is presented based on some changes to the filters used in the prior work by Spencer et al. Along with this, a novel three stage false positive elimination technique is presented to remove non red-lesion candidates detected in the modified filter approach. Firstly, false positive candidates detected elsewhere in the retina are removed. Secondly, falsely found candidates on the blood vessels are removed. Third, candidates found in the optic disc area are removed using our earlier work. To evaluate our method for the detection of red lesions, we examined two sets of fundus images, first set constituting of 94 images obtained from the routine screening at Regional Institute of Ophthalmology, Thiruvananthapuram and the second set of 89 images from DIARETDB1 database. When determining whether an image contains red lesions, our method achieved a sensitivity of 95.6% and specificity of 93.2%. Also, our method detected 78.9% red lesions present in each of the images.