ASCE-XAI: Anatomical Segmentation and Color Enhancement Guided Explainable AI for Pneumonia Detection in Chest Radiographs

Pneumonia is a principal cause of mortality worldwide, and chest radiography is a primary tool for its diagnosis. However, deep learning techniques for pneumonia detection in chest X-rays often operate as "black boxes" without explainability, limiting clinical trust. In this paper, we propose ASCE-XAI, an explainable AI framework that integrates Anatomical Segmentation and Color Enhancement to guide a convolutional neural network (CNN) for pneumonia detection. First, we segment lung fields in the input radiograph to focus the analysis on the relevant anatomical region. Next, we apply contrast-driven color enhancement to the lung region, improving the visibility of pathological features. A CNN is then trained on these processed images to classify pneumonia types, and its predictions are visualized using Gradient-Weighted Class Activation Mapping (Grad-Cam) for explainability. The segmentation and color enhancement steps ensure that the Grad-Cam heatmaps are constrained to lung areas and highlight clinically meaningful features. Experiments on a public chest X-ray dataset demonstrate that our ASCE-XAI approach achieves high pneumonia detection accuracy (around 95% test accuracy) and provides clear, interpretable heatmaps that align with radiological findings. The proposed method offers a promising direction for building more transparent and clinically reliable AI systems for pneumonia and other chest disease detection.