Identification and Classification of Retinal Diseases by Using Deep Learning Models

Abstract

Vision and eye health are crucial for human life; they must be well-preserved to maintain the life of people. Retinal eye diseases for example Diabetic macular edema (DME), Drusen, and Choroidal neovascularization (CNV) conditions are primarily the result of retinal damage, and since the damage to the retina is identified at a late stage, there is nearly no opportunity to reverse the condition and cure it, meaning the patient would likely lose some or all of their vision. Optical Coherence Tomography (OCT) is a powerful scanning technique that uses optical reflection measurements to provide non-invasive cross-sectional imaging of internal biological tissue structures. This will allow ophthalmologists to get a clear view of the posterior part of the eye and diagnose damage to the retina, macula, and optic nerve at an early stage. The proposed work aims to provide a novel model for classification based on deep learning and a free dataset of retinal images obtained from an OCT device is used to automatically classify the various retinal disorders. We demonstrate the architecture of a deep convolutional neural network (CNN), and visual geometry group 16 (VGG-16) compared the performance of pre-trained models and CNN. We suggested CNN architecture achieved 98.3% accuracy, whereas the VGG-16 model achieved 99.28% accuracy.