EDNET-20:

Abstract

In order to optimize clinical management and treatment, it is crucial to accurately diagnose and classify eye diseases. The significant increase in the occurrence of eye diseases and the frequency of visual impairment recently has highlighted the urgent need for enhanced diagnostic techniques and therapies. Deep learning algorithms have shown significant promise in addressing the issue of disease classification. Current diagnostic methods for eye diseases increasingly include advanced medical imaging techniques, including optical coherence tomography angiography (OCTA), optical coherence tomography (OCT), color fundus photography (CFP), and fundus fluorescein angiography (FFA). This study aims to construct a robust framework for accurately classifying eye diseases by using two distinct color fundus photography (CFP) datasets. To classify the diseases this study proposes a novel classifier model called EDNet-20. The EDNet-20 design represents a novel convolutional neural network. The EDNet-20 architecture saw significant enhancements via a rigorous model tweaking of the Convolutional Neural Network (CNN) model using the Mendeley dataset. The study meticulously used data augmentation and pre-processing techniques to enhance the quality and quantity of images in both datasets. The recommended model attained an accuracy of 94.26% on the Mendeley dataset and 91.80% on the private dataset. Compared to eight high-quality transfer learning models, three models that use attention mechanisms, and the basic convolutional neural network (CNN), the recommended model outperformed in terms of accuracy, precision, specificity, recall, and f1 score. These results provide evidence that deep learning technologies may successfully reduce the incidence of vision loss caused by eye diseases. EDNet-20 will be valuable for medical practitioners in enhancing patient outcomes because of its streamlined design and exceptional ability to accurately classify eye diseases.