QBVMD-Based Tri Channel Feature Extraction Approach for Glaucoma Diagnosis

Abstract

Glaucoma is an incurable ocular disorder caused by increased intraocular pressure (IOP) inside the eye. It is the major and second leading cause of blindness. The available techniques are manual and needed experienced experts. Hence, automated diagnosis of glaucoma (DoG) is needed for mass screening, which is more accurate. In this paper, unified quasi-bivariate variational mode decomposition (QBVMD)-based tri channel feature extraction approach (TCFEA) for glaucoma diagnosis from color fundus images (CFIs) is proposed. Initially, CFIs are resized and decomposed into three different color components viz. red component (Rc), green component (Gc), and blue component (Bc). In TCFEA, Rc, Gc, and Bc are passed through three channels for equalization, filtering, decomposition, feature extraction followed by features concatenation to create a features vector set. The features vector sets are further reduced to find the most discriminating features which are fed to the least squares support vector machine (LS-SVM) for the classification of CFIs into glaucoma or healthy class. The measured accuracy, sensitivity, specificity, and F-score are 99.7%, 100%, 99.3%, and 99.7%, respectively, for 10-fold cross-validation (FCV) over the ACRIMA image database. Based on experimental analysis and comparison with available methods, it has been concluded that the proposed method (PM) is more effective with accuracy of 99.7% using 10-FCV strategy. © 2025 IETE.