A Multi-Class Deep Neural Network Framework Driven Automated Classification Of Diabetic Retinopathy Using Retinal Fundus Images

Abstract

Diabetic Retinopathy (DR) is one of the leading causes of vision impairment and blindness among diabetic patients worldwide. Early and accurate detection of DR is essential to prevent severe visual complications through timely medical intervention. This study proposes an automated deep learning-based framework for multi-class classification of Diabetic Retinopathy using retinal fundus images. The proposed model leverages convolutional neural network architecture to effectively extract discriminative features and classify retinal images into five stages: No DR, Mild NPDR, Moderate NPDR, Severe NPDR, and Proliferative Diabetic Retinopathy (PDR). The images were pre-processed through resizing and normalization to enhance model performance. The model was trained using the Adam optimizer with categorical cross-entropy loss and evaluated using standard performance metrics including Accuracy, Precision, Recall, F1-score, and Area Under the ROC Curve (AUC). Experimental results demonstrate that the proposed framework achieved an overall accuracy of 96.4%, precision of 0.962, recall of 0.964, F1-score of 0.963, and an AUC of 0.978, indicating superior classification performance and strong discriminative capability. The findings suggest that the proposed automated system can serve as a reliable and efficient tool for early-stage DR screening and assist ophthalmologists in clinical decision-making. The integration of deep learning techniques into retinal image analysis significantly enhances diagnostic accuracy and reduces manual workload, contributing to improved healthcare outcomes.