Md. Mahi Uddin

Pneumonia is still a big health problem around the world, thus we need quick and accurate ways to diagnose it. While deep learning models have shown promise, their clinical adoption is often hindered by a lack of interpretability and an inability to focus on diagnostically salient regions. To overcome these limitations, this paper introduces PneumoniaXAttnNet, a novel deep learning framework that integrates a custom channel- spatial attention module with a fine-tuned Xception backbone to enhance feature representation by concentrating on critical areas within radiographic images. Our model shows state-of- the-art performance when tested on the public Chest X-Ray Images (Pneumonia) dataset, with an accuracy of 97.35% and an Area Under the Curve (AUC) of 0.9927, which is better than previous benchmarks. Beyond quantitative metrics, we employed a suite of explainable AI (XAI) methods, explicitly Grad-CAM, Simple Gradients, and SmoothGrad, to visually validate that the model’s predictions are grounded in clinically relevant pathological features, such as areas of lung opacity. By combining high diagnostic accuracy with transparent, verifiable decision-making, PneumoniaXAttnNet represents a significant step towards developing a robust and trustworthy computer- aided diagnostic tool for clinical practice. Index Terms—Pneumonia Classification, Deep Learning, Con- volutional Neural Networks (CNN), Attention Mechanism, Ex- plainable Artificial Intelligence (XAI), Chest X-Ray