Supervised pixel classification into arteries and veins of retinal images

With the emerging computation techniques in the field of medical science such as in Ophthalmology; it is often required an automated technique for identification of pathological condition such as diabetic retinopathy which might cause serious problems like blindness. Retinal diseases are often characterized by modification in retinal vessels. Retinal blood vessels observed with fundus imaging provides important indicators not only for clinical diagnosis and treatment of eye diseases but also for systemic diseases such as diabetes, hypertension etc. which manifest themselves in the retina. Quantitative structural analysis of the retinal vasculature not only helps in the diagnosis of retinopathies but also provides potential biomarkers of systemic diseases. Such as arteriole to venule width ratio (AVR) is a parameter indicative of microvascular health and systemic disease. In this paper we performed retinal vessel´s pixel classification into arterioles and venules using Neural Network on DRIVE database. Two types of feed-forward Neural Network are used: Back Propagation Network (BPN) and Probabilistic Neural Network (PNN). BPN gives 83.9% and where as PNN gives 85.1% pixel classification on 20 images. The ROC curve for BPN and PNN has value 0.83 and 0.87 respectively for the DRIVE dataset.