Early identification and grading of Diabetic Retinopathy (DR) is crucial to prevent vision loss. In this study, we used a large dataset of pre-processed retinal scans to assess the performance of deep learning models in achieving this goal. The dataset was processed by combining, enhancing, resizing, cleaning, balancing, and applying Gaussian filtering to retinal images from APTOS 2019 Blindness Detection and DRD. We focused on the ResNetRS models, specifically ResNetRS101, which has been shown to perform exceptionally well in medical imaging tasks, making it a promising tool for DR analysis. Our findings indicate that ResNetRS has excellent feature learning capabilities that can identify subtle indicators that are crucial for diagnosing the condition. Additionally, its deep architecture and residual connections ensure efficient gradient flow, reduce overfitting, and enhance performance across diverse datasets. Our proposed model achieved a training accuracy of 98.9% and a test accuracy of 93% which compete with the related works.