Automated Detection and Classification of Diabetic Retinopathy Using Convolutional Neural Networks: A Deep Learning Approach

Diabetic retinopathy is one of the leading factors in causing visual impairment and even blindness among individuals with diabetes; thus, it requires early detection to enable timely intervention and treatment. In this work, researchers present a CNN-based approach for automatic detection and classification of different DR stages, namely Healthy, Mild DR, Moderate DR, Proliferative DR, and Severe DR. The proposed model trained and tested on the well-curated dataset yields an overall 90% accuracy. From the classification report, it can be observed that the best detection performance comes from the Healthy and Moderate DR stages with F1-scores of 0.97 and 0.81, respectively. However, the model still showed lower recall for Proliferative DR and Severe DR, with F1-scores of 0.63 and 0.57, respectively, indicating challenges in detecting more advanced stages. Despite these challenges, the model's overall performance can be inferred from its macro-average F1-score of 0.75 and weighted-average F1-score of 0.90, which show that it performs well, especially for prevalent classes. This work underlines the potential of using CNNs for automated DR detection. However, class imbalance and poor recall from advanced stages remain two important bottlenecks where further optimization remains necessary. Future work will involve sophisticated data augmentation, ensembling technique, and loss functions that can bring up the accuracy of the model to a level that is of much value in early and reliable diagnosis of DR in clinical practice.