An Integrated Gabor-SMOTE-CNN Framework for Multi-class Diabetic Retinopathy Severity Classification and Automated Screening
Year: 2026
Type: article
Abstract: Diabetic retinopathy (DR) is a leading cause of visual impairment worldwide, requiring early and accurate detection to prevent severe vision loss.However, DR classification remains challenging due to class imbalance and subtle variations in retinal lesions.This study proposes a deep learning-based framework that integrates Gabor feature enhancement, Synthetic Minority Over-sampling Technique (SMOTE), and Convolutional Neural Network (CNN) classification for multi-class detection of DR severity.Experiments were conducted on the Messidor dataset using controlled hyperparameter configurations involving training epochs, batch sizes, and optimization algorithms.The best performance was achieved using 50 training epochs, a batch size of 32, and the RMSprop optimizer.The proposed framework achieved 81.00% accuracy, 79.25% precision, 78.00% recall, and 76.50% F1-score.ROC analysis produced AUC values of 0.93, 1.00, 0.99, and 0.97 for Mild, Moderate, Proliferative DR, and Severe classes, respectively.A web-based prototype was also developed to demonstrate the practical feasibility of the proposed approach for automated diabetic retinopathy screening.
Institutions: Universitas Dian Nuswantoro, Institut Teknologi Garut
Language: English
Topic: Retinal Imaging and Analysis
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7204653261
