Diagnose eyes diseases using deep learning algorithms

Early diagnosis of eye illnesses is the only way to avoid blindness and to guarantee prompt treatment. A crucial part of early eye disease screening is using fundus images. However, as deep learning (DL) offers a precise classification for medical images, using these methods for fundus images makes sense. Recently, DL architectures were applied extensively to image recognition applications. In the presented study, we use DL models, like Convolutional Neural Networks (CNNs), to identify eye diseases in humans. Due to the development of DL methods, investigations on the detection of eye diseases have produced some fascinating results; nevertheless, most of them are restricted to a particular disease. Ocular Disease Intelligent Recognition dataset is used to assess the suggested approach. This has five thousand images representing eight distinct fundus classes. Those classes correspond to various eye diseases. This work will provide an illustration of the five-step recommended system for diagnosing eye problems. The first step of the model is to collect data sets. The second step is to divide the data sets into 70% for training and 30% for testing. The third step is pre-processing to enhance prediction (converting color images into gray-scale, Histogram Equalization, BLUR and resizing process). The fourth step is to use a variety of the feature extraction algorithms (SIFT and GLCM algorithms) are performed to remove redundant information and extract features from the original data, where features are extracted before they are fed into the classifier for the purpose of accelerating the classification process. In the fourth phase, this study created a CNN-based diagnostic system. Next, it features a model for the prediction of presence or absence of diseases in the patient. The findings demonstrated that the classifiers reached highest accuracy of 99.9%. In addition, we see that our best-performing model outperforms several state-of-art techniques in producing competitive outcomes.