Pneumonia Classification using VGG19 Transfer Learning and Data Augmentation with PCA Across Diverse Chest X-Ray Datasets

Abstract

Pneumonia remains a significant global health issue, particularly in low-resource settings, making the development of efficient and accurate diagnostic tools crucial. This study presents a deep learning-based approach using transfer learning with a fine-tuned VGG19 model for pneumonia classification from chest X-ray images. The model was tested across three different datasets (D1, D2, D3) to evaluate its performance in both within-dataset and cross-dataset scenarios. Preprocessing involved resizing the images, followed by data augmentation to improve model generalization. Principal Component Analysis (PCA) was applied to reduce the dimensionality of features extracted from the convolutional layers, optimizing both training time and accuracy. The model achieved high classification accuracies, with performance exceeding 99% in scenarios where the training and testing datasets were identical, as demonstrated by results on D1 and D2. However, cross-dataset evaluation, particularly when training on D1 and testing on D3, revealed a noticeable drop in accuracy, with a minimum of 79.25% in some metrics. The training time ranged from 00:00:15 to 00:13:08, depending on the dataset complexity, demonstrating that PCA.