Early Detection and Classification of Diabetic Retinopathy through Analysis of Retinal Medical Images Using Deep Learning

In this paper, we propose a model based on deep learning (DL) frameworks for the detection and classification of diabetic retinopathy (DR), along with the assessment of its severity. DL algorithms have been applied to fundus photographs, which are images of the back of the eye, to automatically detect and classify signs of DR. These algorithms can analyze retinal images and identify anomalies, including microaneurysms, hemorrhages, exudates, as well as fluid accumulation in the macula, all of which serve as indicators of the presence of diabetic retinopathy. We utilized the APTOS 2019 dataset sourced from Kaggle, comprising high-resolution retinal images processed with Gaussian filters for diabetic retinopathy detection. Experiments were conducted on six different models, including Visual Geometry Group 16 (VGG16), Visual Geometry Group 19 (VGG19), Residual Network 50 (ResNet50), Inception Version 3 (InceptionV3), basic Convolutional Neural Networks (CNN), and Mobile Network Version 2 (MobileNetV2). The results obtained demonstrate high accuracy, suggesting that this model could contribute to expediting and enhancing the diabetic retinopathy diagnostic process. By automating the detection of anomalies in retinal images, the system proposes an intelligent approach that may help alleviate the workload on medical personnel.