Diabetic Retinopathy (DR) is a disease that poses a high risk of vision loss to individuals. In this work, we examine how the two most important approaches in explainable AI (XAI), LIME and SHAP, can be applied to provide local and global interpretations for a DL model built on the VGG19 architecture with an attention mechanism and trained on DR fundus images. We have assembled and annotated a dataset comprising 4000 images each of Mild DR, Moderate DR, Severe DR, Normal DR, and Proliferative diabetic retinopathy (PDR), dividing it into train and test sets. Subsequently, we trained the pre-trained model, VGG19 with a spatial attention mechanism. The spatial attention mechanism in the modified VGG19 model is to selectively focus on important regions of input images by incorporating attention mechanisms operating on feature maps from specific layers, which are then fused with final features to enhance the model’s performance. LIME and SHAP were then applied to the predictions of the model on the test sets. The proposed model with the spatial attention mechanism has achieved an impressive accuracy of 94%.