ResNet-50 Driven Deep Learning Framework For Diabetic Macular Edema Detection In Retinal Fundus Images

Authors: Y. Swathi, B. Haseena Parveen, M. Venkata Lakshmi, U. Harika, N. Sai Vybhavi

Abstract: Diabetic macular edema (DME) is an advanced stage of Diabetic Retinopathy (DR), caused, often resulting in vision impairment or that can lead to vision loss or blindness if not detected early. Timely and accurate detection of DME is critical for preventing vision loss and improving patient outcomes. In this study, a ResNet-50 based on deep learning framework is proposed for the automated detection and grading of DME from retinal fundus images. The model leverages the residual learning architecture of ResNet-50 to capture multi-level hierarchical features, enabling precise identification of pathological regions in the retina. The suggested model was thoroughly tested on several reference retinal datasets and contrasted with the most recent cutting-edge techniques. With impressive results, including 98% Accuracy, 97% Precision, 98% Recall, and 97.5% F1 Score, it outperformed current methods. Excellent discriminative ability between DME-positive and DME-negative cases is confirmed by the ROC curve analysis, and dependable and consistent predictions are highlighted by the confusion matrix. The robustness of the model is confirmed by the training and validation loss curves' steady convergence.

DOI: http://doi.org/10.5281/zenodo.20392019