Enhancing pneumonia diagnosis through pre-processing approaches and advanced AI models: a comparative study and deployment on web and mobile platforms

Pneumonia, often diagnosed through X-ray imaging, is susceptible to diagnostic errors due to subjective interpretation. Machine learning offers a solution by aiding physicians in reducing these errors and saving time. This study employed various convolutional neural network (CNN) models - Basic CNN, VGG16, and EfficientNet - alongside the vision transformer (ViT) for image processing. Pre-processing techniques were applied to augment the dataset and enhance model performance, yielding satisfactory average area under the curve (AUC) scores on the test set. The most accurate and stable models were deployed on both mobile and web applications for physician accessibility. ViT demonstrated exceptional performance, achieving 95% accuracy on training and validation sets, and 97.5% on the test set. VGG16, despite its age, also performed well with 90% accuracy on training and validation sets, and 91.2% on the test set. While VGG16 performed reliably on both platforms, ViT faced challenges in web deployment due to platform disparities.