Retinal vessel segmentation under pathological conditions using supervised machine learning

In this paper we present an automated blood vessel segmentation system algorithm for the retinal images under pathological conditions like Diabetic Retinopathy (DR) using matched filters and supervised classification techniques. Matched filter has been extensively used in the enhancement and segmentation of the retinal blood vessels due to the cross sectional similarity of the vessels to the Gaussian profile. However in addition to the vessel edges the non vessel edges also gives a strong response to the matched filter leading to false detection. Based on the structural and spatial differences between the segmented vessels and the non vessels components, we propose a classification technique using machine learning approach to mask out the false detection due to non vessel structures. The proposed method shows an increased accuracy than the state of the art matched filter techniques especially in the case of vessel segmentation from pathologically affected retinal images.