Classification of retinal diseases from OCT images using deep learning models

Abstract

Retina is an important part of our vision, but it can easily get affected and create various vision problems like vision loss and others. According to the statistics provided by The World Health Organization, it is estimated that globally at least 2.2 billion people suffer from various retinal disorders. It’s important to accurately classify retinal diseases since early detection can help in taking steps for treatment. In this paper, we have classified different types of retinal diseases which are based on OCT images. OCT images were used because they produce a lot of fine-grained retinal images that are useful for diagnosing and monitoring changes to the retina and optic nerve over time. For the classification, we have used Deep learning Models such as CNN models for predicting the accuracy. Moreover, we have proposed a new model for the classification. Our custom model gives an accuracy of 95.05% which is better compared to other pre-trained models. Both DME and DRUSEN class obtained maximum precision that is 97% and Normal class obtained maximum recall which is 98%. Furthermore, we have used Explainable AI (XAI) Techniques with Grad-CAM for better analysis and created a web application for live visualization of result.