Deep Learning-Based Diabetic Retinopathy Detection: A Pretrained Model Approach with Performance Analysis

Diabetic retinopathy (DR) is a major cause of visual impairment globally, and early and precise detection is required to avoid serious complications. In this work, we compare the performances of three state-of-the-art pretrained convolutional neural networks (CNNs) namely MobileNet, Xception, and EfficientNet for binary classification of diabetic retinopathy (positive vs. negative). The models were tested and trained on a retinal fundus images dataset, and their performance was evaluated using various metrics such as classification reports, confusion matrices, receiver operating characteristic (ROC) curves, and precision-recall curves. The outcome shows that the MobileNet performed better compared to the rest of the models, with the best accuracy, sensitivity, and specificity. This work demonstrates the power of light deep learning models in medical image classification and provides a robust solution for automated DR screening.