Özet:

Abstract Glaucoma is by far the most common retinal condition, wherein the intraocular pressure (IOP) on the eye damages the retina. Glaucoma damages ONH, which leads to visual impairment if not addressed. A skilled ophthalmologist checks the course of glaucoma on the retinal area of the eye. This method is time-consuming and inefficient. As a result, this is indeed a legitimate problem that can be addressed by using deep learning algorithms to automatically diagnose glaucoma. In this research work, a novel glaucoma detection model is developed by following five major phases: “(a) pre-processing, (b) ROI identification, (c) feature extraction, (d) feature selection, and (e) glaucoma classification (normal / diseased)”. Initially, the collected retinal images are pre-processed via wiener filtering (to remove noise) and CLAHE (for contrast enhancement). Then, ROI of pre-processed image is selected via Optimized K-Means clustering technique, wherein the centroids of K-means are optimally selected via Dingo with Enhanced Encircling Optimization Model (DEEO). Subsequently, the features inclusive of color feature (Color Histogram and Color Co-occurrence Matrix (CCM)), texture features (Local Binary Pattern-LBP, Median Local Gradient Pattern-MLGP) are extracted from the identified ROI areas. Further, among the selected features, the most relevant features are selected via Dingo with Enhanced Encircling Optimization Model (DEEO).  This DEEO is a conceptual expansion of standard Dingo Optimizer (DOX). Ultimately, using Improved CNN (I-CNN), the classification of OC and OD for healthy and diseased takes place precisely. Finally, a comparative evaluation is undergone to validate the efficiency of the projected model.

Anahtar Kelimeler: