Classification of Retinal OCT Images Using Deep Learning

In recent years, retinal disorders have become a serious public health concern. Retinal disorders develop slowly and without obvious signs. Every year, millions of individuals all around the world are diagnosed with retinal illness. Retinal illnesses manifest themselves in a variety of ways, but the majority of them result in visual problems. Retinal illnesses may damage any portion of your retina, causing visual problems, and some can even lead to blindness. Various retinal illnesses include Diabetic retinopathy, Macular pucker, Glaucoma, Macular hole, Age-related macular degeneration, Drusen, Central serous retinopathy, Macular edema, Vitreous traction, and Optic nerve abnormalities. Millions of light-sensitive cells (rods and cones) and other nerve cells make up the retina, which receives and organizes visual information. To avoid vision deterioration, early identification and treatment are critical. Optical Coherence Tomography (OCT) is a high-resolution diagnostic technology that can analyze and determine the quantitative distinction in diseased retinal layers. OCT - Optical coherence tomography which uses light waves is a non-invasive imaging procedure that takes cross-section photographs of your retina. It obtains a large number of detailed images of the retina, which are valuable for diagnosing and tracking changes in the retina and optic nerve over time. It is crucial in both diagnosing and selecting the appropriate treatment options. The accuracy of traditional approaches for classifying retinal disorders has ranged from 80% to 91%. As a result, a deep learning image identification system based on convolutional neural networks is presented to classify retinal illnesses more precisely and ideally in their early stages. The OCT pictures of the retina are classified into “AMD, CNV, DRUSEN, DMR, DR, MH, CSR, and Normal Eye” using a lightweight Deep neural network. On the Retinal OCT Images dataset, the accuracy achieved in this study with the aid of VGG16 is around 97%. When compared to other methodologies in the literature, it has a high level of accuracy in categorizing the illness.