Ophthalmic image processing for disease detection

Diabetic Retinopathy (DR) is the most common eye complication of diabetes, and it is a leading cause of blindness and visual loss. Therefore, it is crucial to have an automatic and precise DR grading system, as current clinical procedures are still potentially risky and time-consuming. This paper is mainly concerned with the classification of ophthalmic images for eye disease diagnosis. Convolutional Neural Networks (CNNs) are considered for this task. They are trained and tested on public datasets. After the Structured Analysis of the Retina (STARE) and Drishti-GS dataset images are pre-processed and augmented, they are classified using a proposed CNN, and pre-trained ones, including AlexNet, DenseNet, ResNet-50, InceptionV3, and EfficientNet using different tuning hyperparameters. Several performance metrics are used to ensure resilience of the models. Experimental results show that the proposed CNN outperforms the other pre-trained models, achieving a 99.09% accuracy in 6 min with the Adam optimizer, 50 epochs, 0.0001 learning rate, and a batch size of 64 for classifying the STARE dataset, and a 98.7% accuracy for classifying the Drishti-GS dataset.