ANALISIS PERBANDINGAN ARSITEKTUR CNN (RESNET50, VGG16,INCEPTIONV3, DAN EFFICIENTNETB3) UNTUK DETEKSI DINI RETINOPATI DIABETIK PADA CITRA FUNDUS RETINA

Abstract

Diabetic retinopathy is a serious complication of diabetes mellitus that can lead to permanent vision impairment or blindness if not detected early. Therefore, accurate early detection methods are needed to support the diagnostic process. Advances in artificial intelligence, particularly Convolutional Neural Networks (CNN), enable automated analysis of retinal fundus images for detecting diabetic retinopathy. This study aims to analyze and compare the performance of several CNN architectures, namely ResNet50, VGG16, InceptionV3, and EfficientNetB3, for detecting diabetic retinopathy in retinal fundus images. This research uses a computational experimental approach with the APTOS 2019 Blindness Detection dataset that has been resized to 224×224 pixels. The data undergo preprocessing including normalization and data augmentation, and the labels are converted into binary classes (No-DR and DR) to address class imbalance. The models are trained using transfer learning with a data split of 70% training, 15% validation, and 15% testing. Model performance is evaluated using accuracy, precision, recall, F1-score, and ROC-AUC metrics. The results show that all models achieve high performance with accuracy above 96%. EfficientNetB3 produces the best performance with an accuracy of 0.974, an F1-score of 0.974, and a ROC-AUC of 0.993, indicating that it is the most optimal architecture for early detection of diabetic retinopathy.