Related Experiment Video
Updated: May 9, 2026
10:26
A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
Individualized Prediction of Radiation Pneumonitis Using RP-GAN: Leveraging Global Lung Features and Explainable
Yang-Wei Hsieh1,2, Pei-Ju Chao1,3, Yi-Lun Liao1
1Medical Physics and Informatics Laboratory of Electronics Engineering, National Kaohsiung University of Science and Technology, Kaohsiung, Taiwan.
Technology in Cancer Research & Treatment
|May 8, 2026
Summary
This study developed an AI model using deep learning to predict radiation pneumonitis (RP) risk from CT scans. The whole-lung approach showed high accuracy, aiding personalized radiation therapy planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Oncology
- Radiation Oncology
Background:
- Radiation pneumonitis (RP) is a common side effect of thoracic radiation therapy.
- Accurate prediction of RP risk is crucial for optimizing treatment plans and patient outcomes.
- Current methods often lack precision in identifying patients at high risk.
Purpose of the Study:
- To develop an individualized risk prediction model for radiation pneumonitis (RP).
- To utilize unsupervised image feature learning with deep convolutional generative adversarial networks (DCGAN) for automated feature extraction from CT images.
- To enhance radiation therapy planning through precise RP risk assessment.
Main Methods:
- Retrospective analysis of 180 lung cancer patients treated with VMAT.
- Utilized rotation-based augmentation to expand the training dataset.
- Developed an unsupervised feature extraction model (RP-GAN) using whole-lung, V5Gy, and V20Gy CT image inputs, refined with LASSO and ensemble learning (RF, SVM, KNN, XGBoost, LR).
- Validated using 10-fold cross-validation, independent testing, Grad-CAM, and LIME for interpretability.
Main Results:
- The whole-lung input model achieved an AUC of 0.856 and accuracy of 0.861, with a recall of 0.778.
- Models using only V20Gy dose regions showed significantly lower sensitivity (recall of 0.273).
- Explainable AI (XAI) confirmed the model analyzed the entire lung microenvironment, not just the tumor area.
Conclusions:
- The RP-GAN architecture effectively captures subtle lung textural changes for RP risk prediction without manual annotation.
- This AI-driven approach offers a robust method for individualized RP risk assessment.
- The findings support optimizing radiation therapy plans to minimize RP incidence.