Deep learning for identifying and classifying retinal diseases

Vision and eye health are one of the most crucial things in human life, it needs to be preserved to maintain the life of the individuals. Eye diseases such as CNV, DRUSEN, AMD, DME are mainly caused due to the damages of the retina, and since the retina is damaged and discovered at late stages, there is almost no chance to reverse vision and cure it, which means that the patient will lose the power of vision partially and maybe entirely. Optical Coherence Tomography is an advanced scanning device that can perform non-invasive cross-sectional imaging of internal structures in biological tissues by measuring their optical reflections. which will help the ophthalmologists to take a clear look on the back of the eye and determine at early stages the damage caused to the retina, macula, and optic nerve. The aim of this study is to propose a novel classification model based on deep learning and transfer learning to automatically classify the different retinal diseases using retinal images obtained from Optical Coherence Tomography (OCT) device. We propose a deep CNN architecture and compared the obtained results with pretrained models such as Inception V3 and VGG-16, our proposed CNN architecture gave an accuracy of 98.5 % and Inception V3 model gave an accuracy up to 99.27 % on the test set while VGG-16 gave only 53% we modified VGG-16 architecture by adding more convolution layers and regularization terms to obtain a result up to 93.5%.