Generative Adversarial Network (GANS)For Synthetic Medical Image Generation and Augmentation

ABSTRACT Medical imaging datasets, particularly for chest X-rays, often suffer from limited size, class imbalance, and privacy constraints, leading to suboptimal performance in deep learning-based pneumonia detection models. This paper presents a Deep Convolutional Generative Adversarial Network (DCGAN)-based approach to generate synthetic chest X-ray images of the minority class (normal lungs) to augment the training dataset. The publicly available Kaggle Chest X-ray Pneumonia dataset was used, where the minority class was augmented with 2000 synthetic images generated by a trained DCGAN. Two identical Convolutional Neural Network (CNN) classifiers were trained: Model A on real images only and Model B on the augmented dataset (real + synthetic). Experimental results show an improvement in test accuracy from 71.31% (real only) to 74.68% (real + synthetic), demonstrating the effectiveness of GAN-based augmentation in addressing class imbalance and enhancing model generalization. Visual inspection of generated images and GAN training loss curves confirm realistic synthesis. This approach highlights GANs as a valuable tool for overcoming data scarcity in medical image analysis without compromising patient privacy. Keywords: Generative Adversarial Networks, Data Augmentation, Chest X-ray, Pneumonia Detection, Deep Convolutional GAN, Class Imbalance.