Explainable Deep Learning Architecture For Accurate Emphysema Detection In CT Imaging

Authors: Dr. Meghna Utmal, Ragini Gupta, Sonali Gupta, Khushboo Vishwakarma, Vandana Patel

Abstract: Emphysema, a prominent form of chronic obstructive pulmonary disease (COPD), has a substantial impact on lung function and overall patient well-being. Developing accurate and interpretable computer-aided diagnosis (CAD) systems for emphysema using computed tomography (CT) imaging remains challenging, primarily due to the inherent trade-off between predictive performance and model transparency. In this work, we introduce an interpretable deep learning framework for emphysema classification that integrates Convolutional Neural Networks (CNNs) with attention mechanisms and explainable artificial intelligence (XAI) approaches, including Grad-CAM and feature attribution mapping, to highlight critical pathological regions influencing the classification process. The proposed model was trained and validated on a carefully curated dataset of CT images with balanced emphysema severity levels. Experimental results show a classification accuracy of 97.6%, precision of 96.8%, and a root mean square error (RMSE) of 0.203, surpassing recent deep learning approaches by 5–8%. The incorporation of attention-based interpretability enhances clinical transparency by aligning the model’s salient activation regions with radiologist-identified emphysematous areas. This study presents a unified approach that combines explainable deep learning with attention-driven visualization for emphysema detection, achieving high diagnostic performance while improving interpretability and fostering greater clinical trust.

DOI: http://doi.org/10.5281/zenodo.20675710