Deep Learning-Based Automated Screening of Diabetic Retinopathy from Fundus Images

Diabetic retinopathy is a serious eye disease that can cause blindness if it isn't found early. Traditional approaches for identifying this problem rely heavily on experts reviewing retinal scans, which can take a long time and vary from person to person. To solve this problem, we propose a better approach to detecting diabetic retinopathy by utilising Convolutional Neural Networks (CNNs) with a spatial attention mechanism. In this project, we develop a deep learning model using CNNs to analyse retinal images and automatically detect indicators of diabetic retinopathy. Our technique differs from other projects that identify diabetic retinopathy because it uses a spatial attention mechanism within the CNN architecture. The spatial attention mechanism helps the model identify essential image regions, enabling it to detect even subtle indicators of the condition. The goal of this research is to develop automated, efficient screening procedures that ultimately improve patient outcomes by identifying and treating problems quickly. The suggested automated approach could be a useful tool in clinical settings, helping ophthalmologists with large-scale screening programs and reducing the burden on healthcare systems. Future work will concentrate on enhancing model interpretability, integrating multi-modal data, including patient history, and implementing the system in real-world settings. The study highlights the revolutionary impact of artificial intelligence in improving the early identification and treatment of diabetic retinopathy, ultimately leading to superior patient care and a higher quality of life for people with diabetes. The study uses several datasets, including publicly available fundus image sources, to train and test the algorithm. To address class imbalance and broaden the model's generalisation, data augmentation techniques are used. To get high diagnostic accuracy, transfer learning and hyperparameter tuning are used to improve the CNN architecture. We use evaluation criteria such as sensitivity, specificity, and the area under the receiver operating characteristic (ROC) curve to assess how well a model performs.