Comparative Analysis of Visual Explainable AI Techniques for Chest X-ray Classification

This resource presents a comparative study of visual explainable AI methods for pneumonia-focused chest X-ray classification. Using a DenseNet121 model trained on the RSNA Pneumonia dataset, the work benchmarks eight CAM-based explanation techniques: Grad-CAM, Grad-CAM++, HiRes-CAM, Score-CAM, Ablation-CAM, XGrad-CAM, Eigen-CAM, and FullGrad. The evaluation combines interpretability-focused confidence metrics (average confidence drop under occlusion and confidence increase with explanation-guided regions) with practical runtime analysis to assess real-world usability. In addition to qualitative heatmap comparisons for normal and pneumonia cases, the manuscript reports method-wise quantitative results and full system specifications to support reproducibility. The study is intended to guide researchers and practitioners in selecting XAI methods under clinical constraints where explanation fidelity and computational cost must be balanced.