A.I. Based Pneumonia Diagnosis System

Pneumonia continues to be a major cause of death globally, especially in areas with minimal radiological capability. In this paper, we describe an end-to-end AI system for pneumonia diagnosis using deep learning. We use a pre trained DenseNet-121 and fine-tune it on a filtered chest X-ray dataset available at Kaggle for binary pneumonia vs. normal classification. To facilitate accurate and transparent decision making, the framework incorporates Gradient-weighted Class Activation Mapping (Grad-CAM), which exhibits visually prominent clinically relevant areas of the input images. Thorough experimentation demonstrates an average test accuracy of about 89% with well-balanced performance among classes as evidenced by a thorough confusion matrix and classification report. The system is also implemented as a web application on Gradio for real-time clinical decision support. Issues like data imbalance and interpretability of models are addressed in addition to proposed future enhancements and clinical integration.