Related Experiment Video
Updated: Sep 5, 2025
Author Spotlight: Improving Radiation Therapy Access with Radiation Planning Assistant
Published on: October 6, 2023
DR-only Carbon-ion radiotherapy treatment planning via deep learning
Xinyang Zhang1, Pengbo He1, Yazhou Li2
1Institute of Modern Physics, Chinese Academy of Sciences, Lanzhou 730000, China; Key Laboratory of Heavy Ion Radiation Biology and Medicine of Chinese Academy of Sciences, Lanzhou 730000, China; Key Laboratory of Basic Research on Heavy Ion Radiation Application in Medicine, Gansu Province, Lanzhou 730000, China; University of Chinese Academy of Sciences, Beijing 100049, China.
This study shows that using deep learning for digital radiography (DR)-only treatment planning is feasible for carbon ion radiotherapy. This approach enables patient-specific heavy ion therapy planning with high accuracy.
Area of Science:
- Medical Physics
- Radiotherapy Technology
- Artificial Intelligence in Medicine
Background:
- Carbon ion radiotherapy offers precise dose delivery for cancer treatment.
- Accurate patient-specific treatment planning is crucial for optimizing carbon ion therapy outcomes.
- Current planning often relies on computed tomography (CT) scans, necessitating alternative imaging strategies for certain scenarios.
Purpose of the Study:
- To assess the feasibility of a digital radiography (DR)-only treatment planning workflow for carbon ion radiotherapy.
- To develop and validate a deep learning framework for generating patient-specific CT images from DR data.
- To evaluate the accuracy of dose calculations and treatment plan quality using DR-derived CT images compared to conventional CT-based plans.
Main Methods:
- Developed two deep neural networks to correlate DR with digitally reconstructed radiograph (DRR) and DRR with CT images.
- Utilized anthropomorphic phantoms and head-and-neck cancer patients for validation.
- Assessed image similarity using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity (SSIM).
- Performed dose calculations on predicted CT images and compared them to CT-based dose distributions using relative dose differences and 3D gamma analysis (3 mm, 3%).
Main Results:
- The deep learning framework achieved high image similarity metrics (average MAE: 0.007, RMSE: 0.144, PSNR: 37.496, SSIM: 0.973).
- Relative dose differences between predicted CT- and CT-based dose distributions were within 2% for phantoms and ≤4% for patients.
- Average gamma pass-rates exceeded 98% for both phantom and patient-specific plans, indicating excellent agreement.
Conclusions:
- A patient-specific DR-only treatment planning workflow for heavy ion radiotherapy is feasible.
- Deep learning significantly enhances the accuracy and reliability of DR-based treatment planning.
- This approach holds promise for advancing heavy ion radiotherapy by simplifying the planning process and potentially reducing patient burden.
