Classification of Fundus Images Captured using D-Eye Smartphone Retinal Imaging System

Due to the wide range in the accessibility of eye care services, their affordability, and the population’s eye care literacy, cataracts, age-related macular degeneration, glaucoma, diabetic retinopathy, etc., are among the top causes of vision impairment worldwide. To increase the standards of eye examination and to detect the most common vision impairment causes early in a hassle-free manner, real time and sophisticated dataset and efficient classification model is essential. Therefore, this paper emphasizes on the portable smartphone-based fundus image capturing system, D-EYE which acts like a visual digital window. The D-EYE system enables routine eye screening, provides information on visible eye illnesses, and captures images for additional analysis of certain medical situations. With the help of the D-EYE device, an efficient dataset of fundus images of glaucoma is generated and subsequently image enhancement is performed through multiple image pre-processing techniques like image sharpening and masking techniques. The fundus images of glaucoma are cropped to look like high resolution fundus images of the eye using the Super-Resolution Generative Adversarial Network (SRGAN). To categorize fundus images as normal or abnormal, or as glaucoma affected or not, a binary classification model was finally developed. The outcomes show that, with an accuracy of 85%, the classification model has performed at its peak level.