Enhanced Diabetic Retinopathy Grading with Convolutional Neural Networks and Graph Attention Networks

Diabetic Retinopathy (DR) remains one of the most common cause of vision impairment worldwide. The evolution of high definition Colour fundus imaging has greatly enhanced DR diagnosis but manual image analysis is still time-consuming and vulnerable to changes among different clinicians. However, Current deep-learning-based models, especially those based on Convolutional Neural Networks have demonstrated significant potential. findings in diabetic retinopathy classification often exclude the complex Interconnectivity of the deep features relevant for precision. assessment. Our current research is to fill the gap in the existing DR classification model with Graph Attention Networks with the raw convolutional neural network The proposed model is a framework. It contains two main parts: CNN for preliminary feature extraction from DR images and a GAT to capture dynamic interactions between the very deep features. The GAT layer improves the capacity of the model to emphasize Some relevant attributes and factors cause an increase in and correct DR grading. Outputs of CNN and GAT are amalgamated employing a dynamic weighting approach, thereby enabling the model to adaptively alter the effect of each one, relative to Characteristics of the dataset This integration enhances The classification performance based on the strengths of both. CNNs and GATs.