Explainable AI-Driven Deep Learning Model for Transparent and Accurate Pneumonia Detection Using Chest X-Rays

Cynthia Hayat (Krida Wacana Christian University, Department of Information Systems, Center of Optimization System and Machine Learning, Jakarta)
Florensa Rosani Br Purba (Krida Wacana Christian University, Department of Informatics, Jakarta)
Eka Widrian Suradji (Krida Wacana Christian University, Integrated Research Laboratory, Jakarta)

Journal of Information Systems Engineering and Business Intelligence
Vol. 12 No. 3 (2026): October

Background: Pneumonia remains a national health concern with high morbidity and mortality among vulnerable populations. Chest X-ray is the most widely used diagnostic modality due to its accessibility and low cost. However, its interpretation heavily relies on radiologist expertise and often exhibits substantial interobserver variability, potentially leading to inconsistent diagnoses. Although deep learning models have shown remarkable accuracy in medical image analysis, their adoption in clinical practice is limited by the lack of transparency. XAI has therefore emerged as a promising approach to enhance interpretability and clinical trust.  Objective: This study aims to develop a ResNet-50-based explainable deep learning framework for pneumonia detection in CXR by integrating XAI techniques in the form of Grad-CAM and LIME  Methods: A CNN model with a ResNet-50 backbone was trained and tested using a CXR dataset from clinical sources at partner hospitals in Indonesia. Grad-CAM enhances transparency by highlighting relevant lung areas, and LIME offers local feature based explanations by identifying the most important image regions for model decisions. Performance was evaluated using metrics such as accuracy, precision, recall, F1-score, IoU, Dice Score, mean Average Precision (mAP@0.5), and the agreement level with radiologists measured by Cohen’s Kappa test.  Results: The model achieved an accuracy of 93.2%, precision of 92.5%, recall of 94.1%, F1-score of 93.3%, IoU of 78.4%, Dice score of 87.9%, and mAP@0.5 of 89.7%, with an average inference time of 120 ms per image. Grad-CAM and SHAP visualizations showed up to 91% of the area matched the radiologist’s assessment, indicating nearly perfect agreement, with a Kappa value of 0.90.  Conclusion: Combining ResNet-50 with Grad-CAM creates a pneumonia detection model that is accurate, easy to understand, and matches doctors' opinions. This approach could help doctors trust artificial intelligence more and encourage wider use of technology in healthcare, especially in resource limited settings.  Keywords: Pneumonia detection; Chest X-ray; Deep learning; Explainable AI; ResNet-50; Grad-CAM; Medical image analysis