Diabetic Retinopathy (DR) is a severe eye disease in diabetics that can lead to retinal damage and blindness if not properly managed. Traditional DR screening by ophthalmologists is time-consuming, prompting the need for more efficient methods. This project leverages Deep Learning (DL) to automate DR detection using MobileNet, a lightweight model trained on 3,662 high-resolution fundus images from the APTOS dataset hosted on Kaggle. The MobileNet model classifies DR into five stages—zero through four. To enhance performance further, we explore hybrid models combining MobileNet with other architectures: MobileNet + Graph Neural Network (GNN) and MobileNet + Recurrent Neural Network (RNN). The MobileNet + GNN hybrid aims to capture complex spatial relationships within retinal images, potentially improving classification by interpreting structural patterns. The MobileNet + RNN hybrid seeks to address sequential dependencies and temporal patterns, enhancing the model's ability to detect nuanced changes across different DR stages. These hybrid approaches aim to optimize DR detection, improving both accuracy and efficiency.

Keywords - Deep learning, diabetic retinopathy, dataset, MobileNet, Graph Neural Network (GNN), Recurrent Neural Network (RNN).