Automatic detection of eye diseases using image processing and deep learning methods

Computer systems are developing day by day and entering almost every aspect of our lives. The diversification of developed hardware and software also brings innovations in the field of health and leads to increased use in this field. Nowadays, with Artificial Intelligence applications and medical developments increasing day by day, different systems are being developed to detect eye diseases. By looking at patient data obtained through various imaging methods, doctors use their knowledge and experience in their fields of expertise to diagnose and initiate treatment with high success. Human errors may occur from time to time during this diagnosis and diagnosis stage. Nowadays, the increasing number of patients increases the workload of doctors in detecting eye diseases and prolongs the diagnosis time. This lost time can be quite critical when it comes to human health. Short time savings can prevent irreversible diseases. For this reason, the systems to be developed in this field will both save time and eliminate the possibility of human error. In the present study, Deep Learning method, one of the sub-branches of Artificial Intelligence applications, was used on 3358 fundus images available in the Ocular Disease Intelligent Recognition (ODIR) dataset. The images in the dataset are categorized into 8 branches as Normal (N), Diabetes (D), Glaucoma (G), Cataract (C), Age-Related Macular Degeneration (A), Hypertension (H), Pathological Myopia (M) and Other Diseases/Anormalities (O). In the method, many pre-trained models were used to make decisions while making these classifications and the most successful ones were VGG16, ResNet50, ConvNeXtBase and EfficienNetB0. The results show that the proposed method can classify more than 10.000 records in the dataset with an accuracy rate of 93.46%.