Automatic localization and extraction of optic disc using darwinian PSO and morphological operations

Abstract

Locating and earmarking of the optic disc (OD) is a crucial step in the automatic identification of retinal diseases. In advanced stage of proliferative diabetic retinopathy on disk, the delicate blood vessels starts to grow in the disk and hence needs to be clearly identified for better grading of diabetic retinopathy. Furthermore exudates and optic disc share almost same intensity level and may lead to wrong classification if the latter is not identified and removed before the classification. In this paper, we propose an efficient automated system for OD detection and extraction so that exudates are extracted more effectively which will improve overall accuracy in diagnosis of Diabetic Retinopathy. A novel multilevel thresholding optimized by Darwinian Particle swarm Optimization is adopted to detect optic disc in the fundus image. Later Morphological operations are performed to extract the optic disc with precision. The suggested algorithm is tested on four publically available databases like MESSIDOR, DRIVE, and DIARETDB1. Performance of the algorithm is analyzed from the scatter plot. From the scatter plot, it is observed that manually labeled and automatically detected OD centers have a high positive correlation.