Design and evaluation of an automated pneumonia detection system using chest X-ray images: a multi-model comparative study

The difference between pneumonia and COVID-19 in chest radiographic images remains a clinical challenge due to their similar radiographic features. Although deep learning has shown potential in automated diagnosis, most existing studies focus on single models and lack systematic architectural comparisons. In this study, multiple deep convolutional neural networks are evaluated on public chest X-ray datasets for binary pneumonia and multi-class COVID-19 classification tasks. The results indicate that different models show different strengths, with some achieving higher sensitivity and others better specificity. To address these limitations, an ensemble method is proposed to combine the advantages of multiple models. In addition, gradient-based visualization is applied to interpret the model predictions, and the highlighted regions are generally consistent with areas of clinical significance. In general, this work provides a practical framework for model comparison and offers an interpretable approach to potential clinical screening applications.