Pneumonia detection from enhanced chest X-Ray images based on Double SGAN model

Medical imaging plays a crucial role in clinical diagnosis, however, deep learning models often struggle with imbalanced datasets, which has a negative impact on the accuracy and robustness of pneumonia image classification. This study proposes a deep learning based diagnostic system for pneumonia detection using chest X-ray images. Using the pneumonia MNIST dataset, including pediatric lung images. To address the issue of class imbalance and improve generalization ability, we innovatively propose a Double SGAN model. Firstly, apply spectral normalization to all generator and discriminator layers for stable training and improve performance. Secondly, self-attention mechanisms are integrated into the convolutional layers of the generator to better capture complex image features. Finally, the hinge loss function is used during the training process to further improve learning efficiency. On this basis, this study constructed the ResNet18-SA classification model, which embeds spatial attention mechanism in the residual module of the lightweight network ResNet18 to focus on key feature regions related to pneumonia diagnosis and suppress background noise interference. The experimental results show that the ResNet18-SA model outperforms traditional models in terms of accuracy, precision, recall, and F1 score, reaching 95.83%, 95.87%, 95.21%, and 95.52%, respectively.