Advancing Medical Imaging: A Comparative Exploration of Generative Adversarial Networks for Chest X-ray Synthesis

Chest X-ray imaging is essential for diagnosing and treatment planning of individuals with diseases like COVID-19, Tuberculosis, and Pneumonia. However, creating efficient machine-learning algorithms for early detection is made difficult by a lack of labeled medical images. The study investigates the possibility of using Generative Adversarial Networks (GANs) to create synthetic chest X-ray images for various lung diseases in order to address this issue. This study compares three distinct GAN architectures: Progressive Growing of Generative Adversarial Network (ProGAN), Deep Convolutional Generative Adversarial Network (DCGAN), and Auxiliary Classifier Generative Adversarial Network (ACGAN). Acknowledging constraints within existing GAN architectures, the study employs a hybrid methodology. To improve training stability and image quality, this method combines the progressive generation of ProGAN with the Gradient Penalty loss of Wasserstein GAN (WGAN-GP). An InceptionV3 classifier was used to assess the quality of the generated images. On the original dataset, this classifier recorded an accuracy of 91%. The classifier achieved $72.08 \%, 68.62 \%$, and 45.50% accuracy on images generated by ProGAN, DCGAN, and ACGAN, respectively. The study demonstrates GANs potential to assist in alleviating data scarcity, hence improving the precision and effectiveness of machine learning algorithm training.