Segmentation of retinal blood vessel using generalized extreme value probability distribution function(pdf)-based matched filter approach

Retinal vessels' segmentation is challenging to detect blood vessels for diagnosing diseases such as hypertension, diabetes, and glaucoma. Retinal vessel blood segmentation literature shows that various matched filter approaches are a low profile performance of the kernel template of the intensity of vessel profile. A new matched filter based on the generalized extreme value probability distribution function was proposed to overcome this problem. The proposed retinal blood vessel segmentation approach is divided into stages: preprocessing, generalized extreme value probability distribution function (pdf)-matched filter, and postprocessing. In the preprocessing stage, converting color retinal images into a grayscale retinal image using principle component analysis and enhancing the greyscale retinal image using CLAHE followed by Toggle contrast. In the generalized extreme value, pdf is designed as a new matched filter kernel. The experiment is tested to choose the appropriate parameter values for accurate blood vessel extraction to generate the MFR (matched filtered response) image. In postprocessing, the MFR image of the proposed matched filter is applied on an entropy-based optimal threshold to extract binary-segmented retinal blood vessel image, followed by length filtering to remove the artifact and generate the accurate segment blood vessel. The quantitative performance of the proposed approach is to work on five datasets: DRIVE dataset, STARE dataset, HRF Group-HRF Healthy, HRF- Glaucoma, and HRF-Diabetic Retinopathy to measure in terms of average specificity, average sensitivity, average accuracy, and root-mean-square deviation. The results for the DRIVE database are (98.51%, 65.58%, 95.61%), whereas for STARE database (98.35%, 53.14%, 95.04%). The result of HRF-Healthy database are (98.67%, 39.59%, 93.19%), whereas the HRF-Glaucoma database obtained results (99.20%, 27.13%, 94.28%) and the HRF-Diabetic Retinopathy database obtained results (98.74%, 23.62% 93.56%), respectively.