Diabetic retinopathy (DR) is a leading cause of vision loss globally, where early detection is crucial for timely treatment. This paper introduces a deep learning-based framework for classifying referable diabetic retinopathy (RDR) using the EfficientNet-B5 architecture. By leveraging a fine-tuning approach on a pre-trained model, we enable the extraction of intricate and clinically relevant features from retinal images, facilitating accurate classification of both RDR and diabetic macular edema (DME). A linear classifier is applied to the extracted features for final decision-making. Our model was rigorously evaluated on two key tasks of the MICCAI UWF4DR 2024 challenge: RDR identification and DME detection. In the test phase, the proposed method demonstrated competitive performance, achieving 4th place in RDR classification with an AUROC of 0.9937 and 7th place in DME detection with an AUROC of 0.9697, showcasing the effectiveness of the model in detecting diabetic retinal conditions.