Performance analysis of machine learning techniques for glaucoma detection based on textural and intensity features

Abstract

Glaucoma is one of the significant causes of blindness, which covers about 15% to 20% of the total population, so early-stage detection is essential. The proposed methods apply fast fuzzy C-means approach to determine optics-cup-to-disc ratio (CDR) followed by textural based and intensity-based features of the eye. Textural features include local binary pattern, grey-level co-occurrence matrix, and Harlick features, whereas intensity-based features include colour moment and skewness. Machine learning techniques are applied to extract entropy, horizontal, vertical diameter of optics disc/cup, textural based and intensity-based features that classify the image as glaucoma or healthy image and obtained ophthalmologists verify results. Own dataset of 298 retinal images consisting of both healthy and glaucomatous images is used for experimental analysis. In the proposed method, various machine learning techniques like support vector machine (SVM), K-nearest neighbour, and naive Bayes, report 95.5%, 93.3%, and 94.35% accuracy, respectively.