Convolutional Neural Network: Towards Pulmonary Emphysema Detection from Chest X-ray Images

Emphysema is a respiratory condition identified by the deterioration of lung tissue and decreased ability to exchange gases. Early diagnosis of this disease can improve patient’s quality of life and reduce healthcare costs. Chest Xrays (CXR) are crucial tools for diagnosing pulmonary diseases, but accurate and fast diagnosis of emphysema from these images requires specialized expertise. In this research, we employed a DenseNet-121-based Convolutional Neural Network (CNN) for the precise identification of emphysema from chest X-ray images. DenseNet-121 was selected for its capability to address the vanishing gradient problem and effectively utilize features across all layers, leading to superior performance compared to previous CNN architectures. Our model was trained and evaluated on the OpenI database, which was selected for its high-quality labeled images. Data augmentation techniques were employed to improve model robustness. The model achieved an accuracy of over $\mathbf{9 6 \%}$, with a sensitivity of $\mathbf{9 5. 1 \%}$ and a specificity of $\mathbf{9 7. 9 \%}$. These results surpass those of previous studies, including methods like optimized U-Net architectures. The model’s performance and sensitivity demonstrate its potential for reducing the workload of radiologists while improving diagnosis speed and accuracy. Further evaluations confirmed the robustness and sensitivity of the model in handling different levels of disease severity.