Effective Classification of Eye Disease Using Deep Learning Techniques

Abtracst— Sight and vision assist in connecting people to their surroundings. To continue interacting with the world, it is essential to keep your vision. Based on a report from the National Eye Institute states that 7.7 million Americans suffer from diabetic retinopathy. This may increase to 11.3 billion by 2030.There are many different types of retinal diseases. Some can be caused by genes inherited from your parents, while others can be caused by retinal damage that develops throughout your life. Early detection and diagnosis of retinal diseases are significant. Optical Coherence Tomography (OCT) is a sophisticated non-invasive diagnostic scanning technology that shows the retina’s cross section.In this work, we attempted to use deep learning algorithms to solve the problem of classifying various retinal diseases. Instead of handmade methods, deep learning models that automatically learn key features for specific tasks have progressively improved the performance of medical image analysis. This work mainly focuses on four categories namely CNV, DRUSEN, DME and NORMAl. Inorder to perform analysis, the publicly available retinal OCT images are collected and geometric transformation has been performed for the entire dataset. This resulted in the equalized compilation of a dataset which includes approximately 37,746 images. The dataset is fitted against three algorithms which have given us maximum accuracy namely VGG, 4-layer CNN(custom model), 5-layer CNN(custom model. Hyperparameter tuning has been performed to obtain improved accuracy. Parameters involved in tuning include epochs, optimizers, learning rate, etc. This research presents deep learning-models that can detect and diagnose eye disease.