IMPLEMENTASI MODEL CONVOLUTIONAL NEURAL NETWORK RESNET50 PADA PENYAKIT MATA DARI CITRA FUNDUS

Abstrak

Eye conditions, including cataracts, glaucoma, and diabetic retinopathy, are among the leading causes of blindness worldwide. Early detection through retinal fundus image analysis is crucial to prevent further complications. This study designed a classification system for retinal eye disorders using a ResNet50-based convolutional neural network architecture with transfer learning. The dataset, sourced from Kaggle, contains 4,217 fundus images classified into the following categories: normal, cataracts, glaucoma, and diabetic retinopathy. The training process was carried out with a data ratio of 80:10:10 for training, validation, and testing. The evaluation results showed an excellent model performance of 94%, supported by precision, recall, and F1-score values, indicating optimal results, especially for the diabetic retinopathy class. The high performance of the model demonstrates the effectiveness of the ResNet50 architecture in recognizing visual patterns of eye diseases. However, there is still room for improvement in the glaucoma class, which is often misclassified. This architecture shows excellent potential for application as a decision-making tool in ophthalmology to accelerate disease identification and improve eye health services.