Abstract: Diabetic retinopathy (DR) is one of the leading causes of vision loss globally, particularly among individuals with prolonged diabetes. Early detection and timely intervention are critical to preventing severe vision impairment or blindness caused by this condition. However, traditional diagnostic methods rely heavily on manual examination of retinal fundus images by trained ophthalmologists, which is both time-consuming and resource-intensive. In many underserved or rural areas, access to skilled professionals and diagnostic tools is limited, resulting in delayed or missed diagnoses. The need for an automated, scalable, and accurate system for detecting DR stages is paramount. Such a system can significantly reduce the diagnostic burden on healthcare professionals while ensuring timely identification of at-risk patients. The DenseNet169 model, is fine-tuned for DR detection by adding custom classification layers, including a global average pooling layer, dropout for regularization, and a sigmoid-activated dense layer for multilabel classification. This architecture allows the model to capture intricate patterns in retinal images, crucial for detecting subtle variations in DR severity. By leveraging deep learning technologies like DenseNet169 and integrating them into user-friendly platforms like Flask applications, it becomes possible to democratize access to DR screening, improve diagnostic accuracy, and support healthcare providers in managing the growing burden of diabetes-related eye diseases. This study addresses these challenges by proposing a robust and accessible system for DR detection, bridging the gap between advanced technology and practical healthcare solutions.
Keywords: Diabetic Retinopathy (DR) , Deep Learning , DenseNet169 , Retinal Fundus Images, Automated Diagnosis
|
DOI:
10.17148/IJARCCE.2025.14515
