Enhanced Xception-Based Model for Binary Classification of Diabetic Retinopathy in Retinal Fundus Images

Diabetic retinopathy (DR) remains one of the most prevalent vision-threatening diseases today. It is a complication resulting from diabetes, caused by changes in the retinal blood vessels. Early diagnosis is crucial to preventing its progression to advanced stages that can lead to blindness due of damaged blood vessels. In this study, we focused on improving DR detection through advanced image preprocessing techniques, including Gaussian blur, Contrast Limited Adaptive Histogram Equalization (CLAHE), and unsharp masking. Additionally, we employed GridSearchCV to fine-tune key hyperparameters of the Xception model, optimizing its performance. Our model achieved an accuracy of 94.44% on the APTOS 2019 Blindness Detection dataset and 98.64% on the Eye Disease Image dataset, demonstrating strong precision and recall in DR detection. These results highlight the significant role of customized preprocessing and fine-tuning strategies in enhancing the robustness and predictive accuracy of deep learning models for diabetic retinopathy diagnosis. By enabling earlier and more accurate diagnosis, this study contributes to reducing the risk of vision loss and improving patient care in diabetic retinopathy management.