Explainable Multi-Class Thoracic Diseases Classification using a Hybrid CNN–CBAM Feature Fusion Framework

Thoracic diseases such as pneumonia, tuberculosis, COVID-19, and lung opacity pose significant global health challenges, with overlapping clinical manifestations—such as cough, chest pain, and shortness of breath that make timely and accurate diagnosis difficult. Chest X-ray (CXR) imaging remains a widely used diagnostic modality, as it captures critical structural and textural features of the lungs that facilitate disease differentiation. Traditional diagnostic methods and classical machine learning approaches, however, often fall short in handling the complex, high-dimensional nature of medical images. To address this, we propose a hybrid transfer learning framework that leverages MobileNet and VGG16 for feature extraction enhanced with a CBAM attention block to refine discriminative features, alongside focal loss and class weighting.The proposed methodology was evaluated across multi-class scenarios (three-class, four-class, and five-class classification) using multiple datasets, achieving promising results with 98% accuracy in both three- and four-class classification, and 97% accuracy in five-class classification.