Enhancing Pneumonia Image Classification Accuracy for GAN-Generated Data Using DSITL of Customized VGG19 Features with Hybrid Dimensionality Reduction (DSITL-HDR)

Abstract

Pneumonia is a serious respiratory infection and a global health concern. Deep Learning (DL) models, especially Convolutional Neural Networks (CNNs), have improved pneumonia diagnosis using chest X-ray (CXR) images. However, classifying imbalance dataset remains challenging due to overfitting. The proposed study overcome this issue by generating synthetic images using Generative Adversarial Networks (GANs). The Domain-Specific Inductive Transfer Learning (DSITL) of a customized VGG19 features reduced by dimensionality reduction the reduced features are fed in Random Forest (RF) classifier. The results show improved classification accuracy, highlighting the potential of DSITL- hybrid feature for medical imaging.