Automated classification of retinal images into AMD/non-AMD Class—a study using multi-threshold and Gassian-filter enhanced images

Abstract

Eye is an essential sensory organ in human physiological system and disease in eye severely affects the vision system. Age-related-macular-degeneration (AMD) is a common eye disease in elderly people (age > 60 years) and the untreated AMD will cause severe eye illness including permanent vision loss. This research work proposes a machine-learning-system (MLS) to detect the AMD using the fundus-retinal-images. The various phases of the MLS employed in the proposed work is as follows; (i) Image thresholding based on Shannon’ Entropy and Bat Algorithm (BA + SE), (ii) Gaussian-filter (GF) based image smoothening, (iii) feature extraction, (iv) Feature selection using statistical test, and (v) classifier implementation and validation. The image features, such as the gray level co-occurrence matrix and entropies are considered to categorize the retinal image database into AMD/Non-AMD class. The performance of classifiers, such as Naive–Bayes (NB), decision-tree (DT), K-nearest neighbors, random-forest (RF) and support-vector-machine with linear kernel (SVM) are compared and the experimental outcome confirms that, the classification accuracy of SVM is superior (> 93%) compared to other classifiers.