The early detection of Diabetic Retinopathy (DR) is critical to counteract its global devastating impact on vision health. In recent years, the proliferation of medical imaging technologies and enhanced machine learning techniques have propelled the development of automated screening techniques, notably for DR. In this study, we leverage the optimization capabilities of different Nature-Inspired Algorithms (NIAs) to enhance the classification capabilities of traditional Convolutional Neural Networks (CNN) to accurately categorize the different stages of DR and diagnose it from digital fundus images. In our experiments, the proposed hybrid CNN model with Particle Swarm Optimization (PSO) achieves an accuracy of 98.83%, surpassing other existing approaches.