Detection of Pneumonia Using CNN and its Variants

Severe respiratory infections like pneumonia pose a major global health threat, requiring prompt and precise diagnosis for appropriate treatment. Conventional diagnostic techniques are often subjective and time-consuming due to reliance on radiological expertise. This study leverages deep learning to improve pneumonia detection from chest X- rays, evaluating multiple CNN designs such as ResNet50v2, VGG16, MobileNetV2, CheXNet, InceptionV3, and a customized CNN. Models are trained and tested on a large dataset of labelled chest X-rays, employing preprocessing techniques like data augmentation, scaling, and normalization. Transfer learning optimizes the models where applicable. Performance is assessed using recall, accuracy, precision, and F1-score, alongside ROC curve and confusion matrix analysis. Findings indicate that deep learning models significantly enhance pneumonia detection accuracy, with CheXNet and ResNet50v2 outperforming others. This comparative analysis highlights each model's strengths and weaknesses, advancing the development of reliable, automated diagnostic tools for pneumonia and potentially enabling quicker and more precise clinical decision- making.