Automated Glaucoma Diagnosis From Fundus Images Using an Improved 2D-VMD Framework

Glaucoma is a progressive optic neuropathy which leads to permanent loss of vision as the optic nerve is damaged and, in most cases, this type of visual loss is not noticed until late. It is essential to diagnose it early to avoid irreversible blindness, though the traditional approaches like tonometry, visual field testing, and optical coherence tomography is resource-consuming and require the expertise of the specialist. Alternatively, automated detection of glaucoma using retinal fundus images is fast and cost effective. This paper presents a binary classification framework that operates in Two-dimensional Variational Mode Decomposition (2DVMD). The suggested approach operates on the retinal fundus images of two groups, namely, the RIM-ONE collection (505 images) and a publicly accessible dataset by Wheyming Tina Song (1450 images; 899 glaucoma and 551 normal). Resizing and Contrast-Limited Adaptive Histogram Equalization (CLAHE) are used to preprocess the images. Individual images are then decomposed using 2D-VMD, and entropy-based features are obtained. The Student t-test is used in feature selection to determine statistically significant descriptors and a Support Vector Machine (SVM) in classification is used. Experimental evidence shows that the given framework is effective in distinguishing between the cases of glaucoma and the normal ones. On the RIM-ONE dataset, the method attains a precision of 94.71, and on the bigger IEEE Dataport dataset, the method attains a precision of 91.88 hence confirming its strength and generalizability.