COPDNet: An Explainable ResNet50 Model for the Diagnosis of COPD from CXR Images
Agughasi Victor Ikechukwu, S. Murali, B. Honnaraju
Abstract
Chronic Obstructive Pulmonary Disease (COPD) is a prevalent pulmonary condition marked by enduring respiratory symptoms and airflow limitations. Prompt diagnosis is vital for efficacious disease management and enhanced patient outcomes. This study introduces COPDNet, an explainable deep learning model founded on the ResNet50 architecture, tailored specifically for diagnosing COPD using Chest radiographic (CXR) images. COPDNet employs preprocessing techniques such as normalization, resizing, and data augmentation to boost model performance and generalization. To foster interpretability and clinician trust, explainability methods like Layer-wise Relevance Propagation (LRP) or Grad-CAM are integrated, offering visual explanations for model predictions. In addition, transfer learning is utilized by initializing the model with pre-trained weights from extensive datasets, minimizing training time and refining performance. With a recall of 98.9% on the pneumothorax CXR image, the efficacy of COPDNet in diagnosing COPD was validated, potentially leading to improved patient outcomes and supporting clinicians in making well-informed decisions.