Optic Cup segmentation from retinal fundus images using Glowworm Swarm Optimization for glaucoma detection. (July 2020)

• Record Type: Journal Article
• Authors: Pruthi, Jyotika; Khanna, Kavita; Arora, Shaveta
• Is Part Of: Biomedical signal processing and control. Volume 60(2020)
• Publication Date: 2020-07
• Subjects: Glaucoma -- Glowworm Swarm Optimization -- Optic Cup segmentation -- Retinal fundus images; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.102004
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel algorithm for optic cup segmentation has been proposed using Glowworm Swarm Optimization. The algorithm considers the change in the intensity of the pallor region and vessel bend selection together. The algorithm works well with images having low contrast and weak boundaries. Overlapping error has been reduced by 3.2% as compared to state-of-the-art techniques. The algorithm has obtained the accuracy of 100% with DRIVE, 96.56% with DIARETDB1, 98.75% with STARE, 99.87% with DRIONS-DB, 98.61% with RIMONE. Abstract: Glaucoma is one of the diseases that damages the optic nerve of the eye and can result in permanent vision loss. Hence, it becomes essential to detect the disorder at an early stage. Optic cup segmentation from retinal fundus images is an important step for automated glaucoma diagnosis. In this paper, we have presented Glowworm Swarm Optimization algorithm that helps in automated detection of optic cup from retinal fundus images. The glowworms as agents help in the construction of the solutions by making use of the intensity gradient inside the cup region. The exploration capability of glowworms is derived from the adaptive neighbourhood behaviour, thereby making them capable of detecting optic cup region accurately, even in images having weak cup boundaries or low contrast. The proposed algorithm has been implemented and evaluated on RIM-ONE, DRIVE, STARE, DRIONS-DB and DIARETDB1 datasets for qualitative and quantitative analysis. The average overlapping error obtained is 22.1% for DRIONS-DB Database which is minimum as compared to other approaches namely thresholding based, Ellipse fitting and Ant Colony optimization algorithm.