Retinal Damage Detection Using Conv2D Net

Abstract

In today’s world, eye damage has become common among people with issues like glaucoma, CNV, and other retinal diseases. The retinal image is the basic factor for ophthalmologists to diagnose the different damages in the eye. Traditional eye check-ups are done using an ophthalmoscope or retinoscope. These are the traditional methods wherein the diagnosis is based on the observation of the doctor. To capture a high-resolution cross-section of the retina of patients, the Retinal optical coherence Tomography (OCT) imaging technique was used. OCT images are effective in the display and diagnosis of many retinal conditions. Artificial intelligence (AI) has made a major impact in the domain of medical surgery. In this paper, we have used data consisting of OCT images to identify abnormalities occurring in them compared to a healthy normal eye. We have used Deep learning methods to make effective use of OCT images in processing data and training the model to interpret the type of damage illustrated in it and make decisions based on it. We have proposed to implement a Sequential Model in Keras to detect which retinal damage is present in the OCT image and classify them into four classes- CNV, DME, DRUSEN & NORMAL based on its validation accuracy. To improve the accuracy, we have modified the sets of convolutions with a varied number of layers. We have further tried to increase the validation accuracy by the use of the Sequential model, ResNet50 model, and ResNet50V2 model. Out of all the models that have been used, the Resnet50 model proved to give the highest validation accuracy of97%. The given project would be helpful to Ophthalmologists to give fast, reliable, and efficient diagnoses of the eye with would aid them to begin the treatment of that damage at an early stage.