Analisis Performa Arsitektur Inception-ResnetV2 dalam Klasifikasi Tingkat Keparahan Diabetic Retinopathy pada Dataset Messidor-1

Abstract

Diabetic retinopathy is a chronic complication of diabetes mellitus that can cause visual impairment and even permanent blindness. Early detection of the severity of this disease is crucial to prevent further damage. Therefore, this study proposes the use of deep learning technology to assist in the automatic classification of retinal images. This study used the Messidor-1 dataset, which contains retinal fundus images with four severity categories: normal, mild, moderate, and severe/proliferative. In this study, the dataset was divided into training, validation, and testing data in an 80:8:12 ratio. The models used in this study consisted of three deep learning architectures: InceptionV3 and ResNet50 as comparison models, and Inception-ResNetV2 as the main model. The training process was carried out using the Adam optimizer and the categorical crossentropy loss function. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The results showed that the InceptionV3 model achieved an accuracy of 63%, ResNet50 achieved 79%, and Inception-ResNetV2 achieved the best performance with an accuracy of 97%. Therefore, it can be concluded that the combination of the Inception architecture and residual connection can improve feature extraction capabilities and stability during the model training process.