Pneumonia remains a worldwide health issue, and early and accurate diagnosis is necessary for effective treatment. This research explores the application of deep learning in automatic pneumonia detection by comparing four advanced convolutional neural network (CNN) architectures- ResNet, ResNeXt, ResNeSt, and ConvNeXt V2- on chest X-ray images.

Transfer learning was employed to fine-tune and evaluate the models on the dataset. The study compares their classification accuracy, precision, recall, and F1-score to determine their feasibility in applications. While single architectures have been compared in earlier research, a direct comparison of ResNeSt's split-attention to ConvNeXt V2 compound scaling has not been explored.

With the adjustment of class imbalance and optimization of model performance, this study aims to decide upon the most effective deep learning model for the diagnosis of pneumonia and provide insights into AI-driven medical diagnostics and potential integration into healthcare systems.