Eye Disease Classification Using VGG-19 and TensorFlow on Fundus Images

Abstract

Eye disorders are health problems that can reduce visual quality and may lead to blindness if they are not detected at an early stage. The use of deep learning technology has become a promising approach to support the classification of eye diseases based on medical images. This study aims to classify eye diseases from fundus images using a Convolutional Neural Network architecture based on VGG-19 with the support of TensorFlow and Keras frameworks. The dataset used in this study was obtained from Kaggle and consists of eight classes, namely Normal, Cataract, Diabetes, Glaucoma, Hypertension, Myopia, Age Issues, and Other. The research stages include data collection, image preprocessing, dataset generation, training and testing data splitting, VGG-19 model design using transfer learning, model training, model testing, and evaluation using accuracy, precision, recall, F1-score, and confusion matrix. The testing results show that the model achieved an accuracy of 78.24% on the testing data with a loss value of 0.0983. Meanwhile, evaluation using k-fold cross validation obtained an accuracy of 91.35% with a margin of ±2.25%. The confusion matrix results indicate that Cataract and Myopia achieved the best classification performance, while the Normal class still showed a low recall value. These findings indicate that TensorFlow-based VGG-19 can be used for eye disease classification, although further development is still needed to improve the model’s generalization ability