The integration of computer vision into healthcare is transforming medical image analysis, particularly in the classification of lung diseases from chest X-ray (CXR) images. This study evaluates several deep learning architectures-VGG16, ResNet, Inception, DenseNet, and EfficientNet-for classifying CXR images into three categories: Normal, Pneumonia, and COVID-19. The models were trained and validated on a diverse CXR dataset, with particular attention to dataset constraints and imbalanced class distributions, ensuring robust performance across various clinical scenarios. Among the architectures, EfficientNet achieved the highest accuracy in multi-class classification, demonstrating superior performance in identifying lung conditions. The model’s decision-making process was visualized using Gradient-weighted Class Activation Mapping (Grad-CAM) alongside other interpretability techniques, highlighting specific image regions that influenced predictions. These visualizations improve the transparency of the AI system and provide diagnostic insights to assist healthcare professionals in clinical decision-making. Additionally, enhanced visualizations were used to explore ambiguous cases between Pneumonia and COVID-19, offering further clarity into model predictions. This work underscores the importance of explainable AI in medical imaging, paving the way for more trustworthy and effective diagnostic tools.