Rapid detection of diabetic retinopathy in retinal images: a new approach using transfer learning and synthetic minority over-sampling technique

Abstract

The challenge of early detection of diabetic retinopathy (DR), a leading cause of vision loss in working-age individuals in developed nations, was addressed in this study. Current manual analysis of digital color fundus photographs by clinicians, although thorough, suffers from slow result turnaround, delaying necessary treatment. To expedite detection and improve treatment timeliness, a novel automated detection system for DR was developed. This system utilized convolutional neural networks. Visual geometry group 16-layer network (VGG16), a pre-trained deep learning model, for feature extraction from retinal images and the synthetic minority over-sampling technique (SMOTE) to handle class imbalance in the dataset. The system was designed to classify images into five categories: normal, mild DR, moderate DR, severe DR, and proliferative DR (PDR). Assessment of the system using the Kaggle diabetic retinopathy dataset resulted in a promising 93.94% accuracy during the training phase and 88.19% during validation. These results highlight the system's potential to enhance DR diagnosis speed and efficiency, leading to improved patient outcomes. The study concluded that automation and artificial intelligence (AI) could play a significant role in timely and efficient disease detection and management.