Advanced Diabetic Retinopathy Classification Using Context-Gated Dilated Convolutional Network

Diabetic retinopathy (DR) is the leading cause of global vision impairment, requires accurate and early detection. The main challenge in DR classification is improving feature extraction and classification accuracy. To overcome this context, a lightweight feature extraction is required. This paper presents a Context-Gated Dilated Convolution Network (CGDCN) to categorize DR phases from retinal images efficiently. The proposed CGDCN has five gated-dilated convolutional layers that extract complex features, a pooling layer that reduces spatial dimensions while retaining critical features, and a fully connected layer for final classification into five DR stages. Context-aware modules in gated-dilated convolutional layers is used. These modules calculate feature map mean intensity and modify feature extraction paths with gating signals. To enhance the network’s capacity to capture small details and contextual information using the suggested algorithm. Finally, the CGDCN framework uses context-aware gating methods and dilated convolutions to detect DR with high precision. The proposed CGDCN architecture achieved significantly higher accuracy (98.178%) and lower loss (0.060) compared to the ResNet50 model (88.73% accuracy and 1.3003 loss) for diabetic retinopathy classification, demonstrating its superior performance and efficiency