Diabetic Retinopathy Classification Using Densenet121 and Web Application Deployment

Diabetic Retinopathy (DR) is a leading cause of preventable blindness globally, particularly among individuals with prolonged diabetes. Early detection and timely intervention are crucial to mitigate severe vision impairment. This work presents an automated system for DR classification using a deep learning-based approach. A pre-trained DenseNet121 model is fine-tuned for five-class classification of DR severity levels, leveraging techniques such as image resizing, normalization, and augmentation to enhance model robustness. The system is deployed via a Flask-based web application, providing a userfriendly interface that allows users to upload retinal images and obtain real-time predictions with confidence scores. The model's performance is evaluated using metrics such as accuracy, precision, and recall, demonstrating its potential for large-scale DR screening. By integrating advanced deep learning with an accessible deployment platform, this solution aims to assist healthcare professionals in streamlining diagnostics, improving early detection rates, and reducing the burden on ophthalmologists. Future enhancements include expanding the dataset, optimizing model performance, and integrating with electronic health records for broader clinical applicability. The proposed model achieved an overall accuracy of 91.51 %, with precision of $\mathbf{9 3. 1 9 \%}$, recall of $\mathbf{9 3. 2 3 \%}$, and $\mathbf{F 1}$-score of $\mathbf{9 3. 1 1 \%}$, highlighting its effectiveness in multi-class DR classification.