Viral pneumonia detection during the COVID-19 pandemic using deep learning and DCGAN-based data augmentation

Deep learning has shown strong potential for automated chest X-ray analysis in the diagnosis of respiratory diseases. However, accurate detection of viral pneumonia remains challenging due to its radiographic similarity with COVID-19 and limited labeled data. In this study, we aim to detect viral pneumonia in chest X-ray images using CNNs, MobileNet, and an ensemble approach, with DCGAN-based data augmentation to address dataset limitations. A total of 317 real chest X-ray images were used, and 1,500 synthetic images were generated using DCGAN to enhance training diversity. Two experimental setups were conducted: (1) training and testing on original data only, and (2) training on combining real and synthetic data with evaluation on real test images. In the second experiment, the improved CNN achieved 92.4% accuracy, MobileNet reached 96.97%, and the ensemble method attained 96.97%. The results demonstrate that DCGAN-based augmentation combined with ensemble learning improves classification performance and robustness. The proposed approach provides an effective tool for supporting automated viral pneumonia detection from chest X-ray images during the COVID-19 pandemic.