Deep Learning-Based Optimized Automated Detection of Diabetic Retinopathy Using Retinal Fundus Images

Abstract

Diabetic Retinopathy (DR) is one of the leading causes of vision impairment and blindness worldwide, often progressing without noticeable symptoms in its early stages. Early and accurate detection is crucial for timely intervention and preventing irreversible damage. This study presents a deep learning-based framework for the automated detection of DR using retinal fundus images. The proposed approach leverages a Convolutional Neural Network (CNN) architecture optimized to extract discriminative features from high-resolution fundus images, enabling robust classification of DR stages. A large-scale, publicly available DR dataset was pre-processed through contrast enhancement, noise reduction, and data augmentation techniques to improve model generalization and performance. The model was trained and validated using stratified data splitting, with accuracy, sensitivity, specificity, precision, and F1-score serving as key evaluation metrics. Experimental results demonstrate that the proposed model achieves superior performance compared to conventional machine learning methods and baseline CNN architectures, attaining high accuracy and balanced sensitivity–specificity scores. The system successfully identifies subtle retinal lesions such as microaneurysms, hemorrhages, and exudates, which are critical indicators of DR severity. This work highlights the potential of deep learning as a non-invasive, cost-effective, and scalable solution for DR screening in both clinical and remote healthcare settings. The proposed framework can serve as a decision-support tool for ophthalmologists, thereby improving diagnostic efficiency and contributing to the reduction of vision loss caused by diabetic complications.