Diagnosis of retinal diseases using optical coherence tomography images and convolutional neural network

The retina is a light-sensitive layer that enables vision and perception of colors. Disturbances in the retina negatively affect people's quality of life. With the advancement of technology in the treatment of retinal diseases, the use of computer-aided diagnostic systems has become widespread. Disturbances occuring in the retina can lead to serious problems, including blindness and permanent damage to the retina. Retinal diseases can result in sudden or gradual vision loss, blurred vision, dark and blotchy images in the field of view. Early diagnosis and treatment of retinal diseases are of great importance. Retinal tomography images are used by ophthalmologists to diagnose retinal diseases and prevent possible permanent damage in the eye. Thus, early diagnosis facilitates treatment and ensures better clinical outcomes. In this study, a deep learning model using Convolutional Neural Networks (CNNs) is proposed for the early detection of retinal diseases and the application of appropriate treatment methods. The proposed model is applied on the ODIR-5K dataset. The ODIR-5K dataset contains retinal images of eight different types of diseases, including Myopia, Hypertension, Glaucoma, Age-related Macular Degeneration (ARMD), Diabetic Retinopathy (DR), Cataract, Other abnormalities, and normal individuals. The images in the ODIR-5K dataset were first equalized to have consistent resolution and dimensions, and then underwent preprocessing steps including histogram equalization using the clahe method. Scaling operations were also performed. Furthermore, the accuracy rates of transfer learning methods were obtained as follows: VGG19 (98.16%), ResNet50 (98.86%), Xception (95.36%), InceptionV3 (98.2%) and the proposed convolutional neural network model (97.51%). The proposed model achieved better classification results for images of cataracts and normal retinas compared to pre-trained transfer learning methods. The proposed deep learning-based CNN model can be used for the early detection of retinal diseases.