The early and accurate detection of lung diseases such as COVID-19, pneumonia, tuberculosis, and others remains a critical challenge in global healthcare, particularly in low-resource settings where timely diagnosis can significantly impact patient outcomes. This study introduces a deep learning-based multiclass classification framework leveraging chest X-ray images to identify five distinct lung conditions, including a “Normal” category to address ambiguous cases and enhance diagnostic reliability. Multiple convolutional neural network architectures, including custom CNNs, ResNet18, ResNet50, and MobileNetV2, were rigorously evaluated, with ResNet50 emerging as the top performer, achieving an accuracy of 99.18%, precision of 98.28%, and F1-score of 98.09%. The dataset was meticulously preprocessed and balanced using advanced augmentation and oversampling techniques to ensure robust and generalizable performance across diverse clinical scenarios. Evaluation metrics, including precision, recall, F1-score, and ROC–AUC curves, were employed to validate the models’ effectiveness, with ROC curves demonstrating near-perfect discrimination. Interpretability was enhanced through Grad-CAM, which visualized clinically relevant features learned by the models, aligning with established radiological patterns. This approach surpasses previous benchmarks and exhibits significant potential for real-world deployment in automated radiology screening systems, particularly in resource-constrained environments.