Eye Disease Classification Based on Fundus Images Using Convolutional Neural Network

The eye is a vital organ in the human body. Through the eyes, humans can absorb over 80% of visual information used to carry out various daily activities. However, in Indonesia, there are many cases of vision impairments that, if not properly addressed, can lead to blindness. Examples of such impairments include cataracts, glaucoma, and diabetic retinopathy. Currently, there are many research studies on eye diseases that are aided by technologies like Artificial Intelligence (AI). One of the AI technologies that is commonly developed is Convolutional Neural Network (CNN). CNN is a deep learning framework that excels in solving image or picture classification problems because its architecture applies convolution layers, which can break down images to extract features and easily reduce the high dimensionality without losing the image’s information itself. In this study, the author used various CNN models to find the most suitable model for classifying eye diseases, including EfficientNet, ResNet, Inception V3, and many others. Through various model training experiments, the accuracy for the ResNet50 model was found to be 95% with a loss of 17%, followed by Xception with an accuracy of 95% and a loss of 21%, and EfficientNetV2B2 with an accuracy of 92% and a loss of 25%.