Code for RP-GAN Radiation Pneumonitis Prediction

This repository contains the implementation of the RP-GAN framework for radiation pneumonitis prediction. The proposed pipeline includes:(1) GAN-based feature extraction using a DCGAN architecture,(2) LASSO feature selection with 10-fold cross-validation, and(3) stacking ensemble learning for outcome prediction. The discriminator of the RP-GAN is used to extract multi-layer feature representations, which are subsequently aggregated and used for downstream modeling. Model performance is evaluated using multiple metrics, including AUC, accuracy, precision, recall, F1-score, and calibration analysis. This repository provides a modular and reproducible implementation of the complete workflow. Due to patient privacy and institutional review board (IRB) restrictions, raw medical imaging data are not publicly available.