The classification and detection of COVID-19 from chest X-ray images using deep learning models is challenging due to the limited amount of training data and the potential for overfitting. Data augmentation remains one of the most effective solutions to address these issues. However, traditional augmentation techniques often generate images that lack the visual characteristics that are most informative for the model. In this paper, we propose a data augmentation approach based on explainable deep learning for COVID-19 diagnosis using chest X-ray images. Our method leverages explainable deep learning techniques to identify the most significant regions in chest X-ray (CXR) images, ensuring that the model focuses on critical areas for disease detection. We extract saliency maps, which act as an importance filter derived from explainability methods. These saliency maps are then used to guide the data augmentation process, ensuring that the generated images emphasize the features most relevant for accurate diagnosis. This method enhances the learning process of complex models by providing a more informative training dataset. It can result in a significant increase in accuracy (approximately 3–20%), a reduction in false-positive images, and a minimization of loss functions and overfitting issues.