Add Example: Adversarial Pneumonia Classification (CXR)

[Ansh187-lab]

Summary

This PR adds a new example to PyHealth under examples/cxr_adv_reproduction/ that reproduces and evaluates the paper:
"An Adversarial Approach for the Robust Classification of Pneumonia from Chest Radiographs"

Files Added

• cxrdataset.py: Custom NIH dataset class
• train.py: Script for training standard and adversarial models
• test.py: Evaluation script with AUROC scoring
• models.py: Contains DenseNet121-based classifiers
• README.md: Detailed instructions on how to run the code

Contribution Type

✅ New example / use case of PyHealth

Notes

• All scripts are runnable in Google Colab
• Uses ~1000 images subset of NIH ChestX-ray14 dataset
• Reproduced results show AUROC of 0.8518 (standard) and 0.8585 (adversarial)