Abstract:

Abstract Diabetic Retinopathy (DR) is a visual condition that occurs because of chronic diabetes, causing damage to retinal blood vessels. It is considered a major cause of vision loss, affecting over 158 million people worldwide. Early detection and diagnosis can minimize the impact on vision and improve the quality of life for those with DR. Therefore, an automated DR detection method is essential. Deep learning models can automate classification and feature extraction, but developing such models from scratch requires a larger and well-annotated dataset. The main challenge in using deep learning for medical image analysis is the lack of annotated training datasets. To address this issue pre-trained deep learning networks are mainly used by the researchers. This study utilizes fine-tuned pre-trained deep learning techniques to derive attributes from fundus images, improving the accuracy of DR classification. To address data imbalances and insufficiencies, various techniques were used to enhance data for each DR class. The results of this study, evaluated using test data, show that our approach outperforms previous techniques in terms of accuracy.

Keywords: