A novel methodology for vessel extraction from retinal fundus image and detection of neovascularization

摘要

Vessel extraction from the retinal fundus images plays a significant role in ophthalmologic disease diagnosis. Proliferative Diabetic Retinopathy (PDR) is the ultimate stage of Diabetic Retinopathy where proliferation of new and fragile blood vessels grow in human retina. These new blood vessels often show a tendency to rupture which further leads to severe damage of human eye. Neovascularization at the disk (NVD) and elsewhere (NVE) are the two general categories of PDR. So, disease diagnosis at the early stage by detecting the newly generated thin vessels demands utmost importance. Literature witness that most of the existing works emphasised on detecting only NVD. The goal of this work is to detect NVD along with NVE, as both the stages are equally devastating. The disease detection requires the extraction of vessels for subsequent analysis. A novel vessel extraction methodology has been proposed here which is capable of extracting the thick and thin vessels for further analysis. The experimental results have been tested and verified with two publicly available datasets of retinal fundus images, DRIVE and STARE. Finally, experiment for NVD and NVE detection has been carried out with DIARET-DBl data-set. Comparison of performance with some other state-of-the-works shows superiority of the proposed methodology.