Reproduce “GAN-based Data Augmentation for Chest X-ray Classification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shobhita Sundaram, Neha Hulkund
Researchers at MIT showed that Generative Adversarial Network (GAN)-based data augmentation improves chest X-ray classification performance, particularly for underrepresented pathologies in low-data regimens. This approach yielded AUC gains up to 0.07 for Fracture and mitigated overfitting, proving beneficial for data-scarce medical imaging.
Key insights
- GAN-based data augmentation offers significant performance gains and overfitting reduction in low-data training regimens, with diminishing returns as real dataset size increases, except for extremely imbalanced classes.
- Standard augmentation techniques may sometimes reduce performance in larger medical datasets due to anatomical non-invariance.
- Models trained with GAN-augmented data develop generalizable features, as evidenced by similar and clinically relevant Class Activation Maps (CAMs) compared to real-data-trained models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.