LAUSR

Abstract Variations in blood vasculature morphology of retinal fundus images is one of the dominant characteristic for the early detection and analysis of retinal abnormalities. Therefore the accurate interpretation of blood vasculature is useful for ophthalmologists to diagnose patients that suffer from retinal abnormalities. A generalized method to detect and segment blood vasculature using retinal fundus images has been proposed in this work using (i) preprocessing for quality improvement of retinal fundus images, (ii) initial segmentation of vasculature map to find vascular and non vascular structures, (iii) extraction of relevant set of geometrical based features from the vasculature map and intensity based features from original retinal fundus image that differentiate vascular and non vascular structures efficiently, (iv) supervised classification of vascular and non vascular structures using the extracted features, and (v) joining of candidate vascular structures to create connectivity. The proposed method is evaluated on clinically acquired dataset and different publically available standard datasets such as DRIVE, STARE, ARIA and HRF. The clinically acquired dataset consists of 468 retinal fundus images comprising of healthy images, images with mild, intermediate and severe pathologies. Test results of the proposed method shows average sensitivity/specificity/accuracy of 85.43/97.94/95.45 on the 785 retinal fundus images. The proposed method shows an improvement of 14.01% in sensitivity without degrading specificity and accuracy in comparison to the recently published methods.