Explainable Pneumonia Detection from Chest X-Ray Images Using a ResNet-50 Based Deep Learning Framework

Abstract

Pneumonia is a serious lung infection responsible for approximately 2 million deaths in children annually and 50,000 adult deaths each year, placing a significant burden on global healthcare. Early and accurate detection is critical, particularly in resource-limited settings. This paper proposes an automated pneumonia detection framework based on the ResNet-50 deep learning architecture applied to chest X-ray images. The model was trained on the publicly available Chest X-Ray Pneumonia dataset (Kaggle) using transfer learning, with 70% of data used for training and 30% for testing. To address model interpretability, a critical requirement for clinical adoption of Gradient-weighted Class Activation Mapping (Grad-CAM) was integrated to visually highlight lung regions influencing classification decisions. The proposed model achieved an overall accuracy of 81.9%, a recall (sensitivity) of 96.7%, a precision of 79.0%, a specificity of 57.3%, and an F1-score of 87.0% on the test set. The high recall is prioritized given the clinical importance of minimizing false negatives in pneumonia screening. Grad-CAM visualizations confirm that the model focuses on anatomically relevant lung regions. This system is intended as a decision-support tool to assist clinicians rather than replace them.