Related Experiment Video
Updated: May 13, 2026
08:25
Radiation Planning Assistant - A Streamlined, Fully Automated Radiotherapy Treatment Planning System
Published on: April 11, 2018
15.3K
Deep reinforcement learning in radiation therapy planning optimization: A comprehensive review
Can Li1, Yuqi Guo1, Xinyan Lin2
1Institute of Operations Research and Information Engineering, Beijing University of Technology, Beijing 100124, PR China.
Summary
Deep reinforcement learning (DRL) shows promise for automating radiation therapy planning. However, clinical application is limited by inefficiency, quality assessment, and interpretability challenges, requiring further research.
Area of Science:
- Medical Physics
- Artificial Intelligence
- Radiotherapy
Background:
- Radiation therapy plan optimization is complex and time-consuming.
- Automated optimization methods are urgently needed.
- Deep reinforcement learning (DRL) offers potential for automation.
Purpose of the Study:
- Review the current state of DRL applications in radiotherapy planning.
- Evaluate the effectiveness of DRL methods.
- Identify challenges and future directions for DRL in radiotherapy.
Main Methods:
- Systematic literature search across major academic databases (Google Scholar, PubMed, IEEE Xplore, Scopus).
- Keywords included "deep reinforcement learning", "radiation therapy", and "treatment planning".
- Data synthesis for critical analysis and overview.
Main Results:
- DRL applications in radiation therapy planning include optimizing treatment parameters, machine parameters, and adaptive radiotherapy.
- DRL has been applied to various cancers (cervical, prostate, lung) and radiation techniques (IMRT, VMAT, SBRT, etc.).
Conclusions:
- DRL significantly advances automated radiation therapy plan optimization.
- Clinical implementation is hindered by inefficiency, limited quality assessment, and poor interpretability.
- Future research should focus on evaluators, parallelized training, and continuous action spaces.
