A new approach for retinal vessel differentiation using binary particle swarm optimization. Issue 5 (3rd September 2021)

• Record Type: Journal Article
• Authors: Irshad, Samra; Yin, Xiaoxia; Zhang, Yanchun
• Is Part Of: Computer methods in biomechanics and biomedical engineering. Volume 9:Issue 5(2021)
• Page Start: 510
• Page End: 522
• Publication Date: 2021-09-03
• Subjects: Retinal vessel classification -- fundus image analysis -- feature selection -- hypertensive retinopathy; Imaging systems in biology -- Periodicals; Imaging systems in medicine -- Periodicals; Biomechanics -- Data processing -- Periodicals; Biomedical engineering -- Periodicals; 616.0757
• DOI: 10.1080/21681163.2020.1870001
• Languages: English
• ISSNs: 2168-1163

Abstract

ABSTRACT: Hypertensive Retinopathy is an ocular disease that occurs due to the presence of hypertension in the body. Hypertensive Retinopathy is quantified using a parameter known as Arteriovenous Ratio. To calculate Arteriovenous Ratio, retinal blood vessels in fundus images are segmented, classified, and measured. The complex structure of retinal vessels and uneven illumination in fundus images make retinal vessel classification a difficult task. In this paper, we propose a method for an improved differentiation of retinal vessels, that employs Binary Particle Swarm Optimization to select the optimal retinal vessel features. We design an objective function that considers the size and relevancy of the feature subset in classifying the retinal vessels. The designed objective function ensures the selected feature subset contains the minimum number of features and it corresponds to maximum vessel classification accuracy. Once, the optimization process has reached the stopping criteria, the selected feature subset is evaluated using Support Vector Machines. We compare the retinal vessel classification accuracy obtained using the proposed framework with the existing state-of-the-art approaches and found our method more accurate and robust on healthy as well as diseased images. The effectiveness of the proposed method is validated using the public as well as a private dataset.