Development of a CNN-based decision support system for lung disease diagnosis using chest radiographs

Chest radiographs, or chest X-rays (CXRs), are widely used as first-line diagnostic tools for detecting various chest diseases. However, accurately interpreting CXRs remains challenging, as human diagnostic performance is influenced by individual expertise and other factors, often resulting in delays, high costs, and potential misinterpretations. To address these limitations, automated computer-based detection systems offer the potential to enhance diagnostic accuracy, reduce costs, and enable timely disease identification. This study presents CXRNet, a novel, efficient convolutional neural network (CNN)-based framework designed for multi-class classification of common chest diseases, including cardiomegaly, COVID-19, pneumonia, tuberculosis, and normal. The proposed CXRNet is a 16-layer architecture trained on frontal CXR images collected from diverse sources to ensure robust generalization across datasets. The model incorporates advanced strategies to overcome the limitations of previous approaches. Extensive testing under three different data distribution conditions demonstrated the model’s superior performance, achieving an average accuracy of 95.7%, precision of 95.3%, recall of 95.3%, and an F1-score of 95.3% for multi-class classification. Furthermore, for binary classification tasks, CXRNet achieved over 98% average accuracy across all conditions, outperforming existing methods. These results highlight the potential of CXRNet as a reliable decision support system for efficient and accurate chest disease diagnosis, paving the way for real-time clinical applications.