Pneumonia Classification Using VGG19 Transfer Learning And Data Augmentation Across Diverse Chest X-Ray Datasets

Abstract

Image classification is widely applied across various fields, yet its application to medical images remains limited. Recently, significant advances in Deep Learning (DL) have sparked increased research interest in medical image recognition. Traditional diagnostic methods, such as Chest X-rays (CXR), often require expert interpretation and are prone to human error. However, training DL models from scratch demands large amounts of labeled data, which are often scarce in the medical domain. This study explores the application of Transfer Learning (TL) using the VGG19 Convolutional Neural Network (CNN) architecture for pneumonia classification across three distinct CXR image datasets (D1, D2, and D3). Data augmentation techniques, including horizontal flipping, rotation, zoom, and contrast adjustment, were employed to artificially expand the training datasets, enhancing model generalization. The study aims to evaluate the effectiveness of VGG19 when pre-trained on a large-scale image dataset and fine-tuned on each smaller, domain-specific dataset. TL is used to leverage features learned from the pre-trained VGG19 model, improving classification accuracy on the target pneumonia datasets, ultimately enhancing patient outcomes and reducing diagnostic errors.