A Progressive Conditional GAN Framework with Multi-Scale Discriminators for Disease-Aware Chest X-Ray Synthesis

DOI

10.2991/978-94-6239-756-9_11How to use a DOI?

Keywords

Generative Adversarial Networks; Conditional GAN; Chest X-ray Synthesis; COVID-19 Radiography Database; Medical Image Aug- mentation

Abstract

The scarcity of large, well-annotated medical imaging datasets poses a significant challenge to developing robust and generalizable diagnostic models. Although chest X-rays (CXRs) are among the most widely used imaging modalities, publicly available datasets often suffer from severe class imbalance and limited diversity, particularly for rare pathologies such as COVID-19 and viral pneumonia. To address these limitations, this study proposes a Progressive Conditional Generative Adversarial Network (PC-GAN) framework capable of synthesizing high-quality, disease-aware, and anatomically realistic CXR images. The generator employs progressive upsampling and residual skip connections conditioned on disease labels to enable class-specific synthesis, while a multi-scale PatchGAN discriminator ensures both local texture realism and global structural coherence. Furthermore, the training process is stabilized using a composite loss formulation that integrates Fourier-band curriculum loss, edge-preserving Sobel loss, auxiliary classification loss, and hinge adversarial loss with differentiable augmentation. The experiments in this study were carried out using the publicly available COVID-19 Radiography Database from Kaggle, which contains four classes named as Normal, Lung Opacity, Viral Pneumonia and COVID- 19, with corresponding image counts of 10192, 6012, 1345 and 3616 respectively. The experiment demonstrates the superiority of the proposed approach. Unlike existing GANs, this framework integrates conditional synthesis, multi-scale discriminators, and composite loss functions, ensuring clinical interpretability and improved anatomical fidelity.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Garima Singh
AU  - Juhi Sonkar
AU  - Ravi Shankar Singh
PY  - 2026
DA  - 2026/08/31
TI  - A Progressive Conditional GAN Framework with Multi-Scale Discriminators for Disease-Aware Chest X-Ray Synthesis
BT  - Proceedings of the Conference on Bioengineering for Global Health (SYMRESEARCH 2.0 2025)
PB  - Atlantis Press
SP  - 135
EP  - 145
SN  - 2468-5747
UR  - https://doi.org/10.2991/978-94-6239-756-9_11
DO  - 10.2991/978-94-6239-756-9_11
ID  - Singh2026
ER  -