Abstract
The retinal blood vessels act as a diagnostic indicator of several severe pathologies like Diabetes, Hypertension, and Stroke. Fundus photograph is the most common modality for examination of eyes. Medical experts use fundus examination during eye diagnosis. However, the quality as well as resolution of fundus images are rapidly increasing, resulting in many challenges. In this paper, an unsupervised automated framework for the extraction of retinal vessels by using enhanced filtering and hessian-based method with hysteresis thresholding is presented. In this framework, we have considered two publicly available datasets, DRIVE and DR-HAGIS consisting of more pathological high-resolution images for experimentation. In this framework, there are four major phases: (i) Pre-processing phase for the extraction of green channel and contrast enhancement; (ii) Vessel enhancement phase for enhancement of wide and thin vessels; (iii) Vessel detection phase for the extraction of vessels; (iv) Post-processing phase for filling the holes and to remove isolated pixels. Our method achieves 0.9601 accuracy on DRIVE dataset and 0.9484, 0.9496, 0.9487, 0.9480 on DR-HAGIS with Glaucoma, Hypertension, Diabetic Retinopathy, Age-Related Macular Degeneration (AMD) images respectively. Our results manifest that the proposed framework performs fine at extracting blood vessels from pathological images in addition to healthy retinal images and is comparable with existing methods.