An Ensemble Learning Framework For Automated Staging Of Diabetic Retinopathy Using Fundus Images

Abstract

Diabetic Retinopathy (DR), a progressive complication of diabetes mellitus, remains one of the leading causes of vision loss worldwide. Accurate and early staging of DR is essential for effective clinical intervention. This paper proposes an ensemble-based classification framework for the automated analysis of DR risk factors and stage prediction using fundus images. The proposed system includes a multi-phase pipeline consisting of preprocessing, segmentation, feature extraction, and classification. In the preprocessing phase, image quality is enhanced using a combination of Bilateral Filtering and Non-Local Means (NLM) Denoising, effectively reducing noise while preserving critical retinal structures. Segmentation is performed to isolate blood vessels and other key anatomical features. Deep features are extracted using a deep residual autoencoder, capturing complex spatial patterns associated with DR progression. For classification, an ensemble of machine learning models—Extreme Gradient Boosting (XGBoost), Radial Basis Function Support Vector Machine (RBF-SVM), and Random Forest (RF)—is employed. This hybrid approach leverages the individual strengths of each classifier to improve generalization and accuracy. Experimental results demonstrate the effectiveness of the proposed framework in distinguishing between different DR stages, with promising performance in terms of accuracy, precision, and recall. The model shows potential for use in clinical decision support systems and tele-ophthalmology platforms.