ChestACGAN: Auxiliary Classifier Generative Adversarial Network for Chest X-ray Images

Computer vision is becoming very popular in the medical sector due to data collection remains costly and data availability is limited. To solve this issue, a generative adversarial network can be used as a sustainable solution, given its ability to generate realistic images that can be used as samples in medical sector and introducing significant variation in the generated dataset. Consequently, this paper suggests employing a GAN-based data generation technique to handle the constraint of limited data availability, alongside custom CNN model for detecting pneumonia from chest x-ray images. In our paper, auxiliary classifier GAN is used to extend dataset by enabling the management of labels for generated images. ACGAN generate high-quality synthetic images identical to actual images, and then real and synthetic images were concatenated and fed into a custom-designed convolutional neural network model to validate the efficacy of data generation. Our research findings show that average classification accuracy increased significantly from $\mathbf{9 2 . 3} \%$ to $\mathbf{9 9 . 4 6 \%}$ after expanding the dataset with generated data. This study is anticipated to have a substantial impact on the medical sector, with the generation of realistic images that can be beneficial to doctors and experts.