Meta-optimization for fully automated radiation therapy treatment planning

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Biology

Background:

• Radiation therapy treatment planning is a complex, time-consuming process.
• Manual adjustments of hyperparameters are iterative and labor-intensive.
• Automating this process can significantly improve efficiency and consistency.

Purpose of the Study:

• To introduce MetaPlanner (MP), a novel meta-optimization framework for automated treatment planning.
• To develop an algorithm that mimics clinical decision-making for treatment plan optimization.
• To provide a publicly available, open-source solution for automated radiotherapy planning.

Main Methods:

• MP employs a derivative-free meta-optimization approach using parallel Nelder-Mead simplex search.
• Hyperparameter weight configurations are optimized to minimize a meta-scoring function.
• The meta-scoring function incorporates clinical considerations like dose homogeneity, conformity, spillage, and organ-at-risk sparing.

Main Results:

• MP was evaluated on 21 prostate and 6 head and neck cancer cases.
• The framework was applied to both Intensity-Modulated Radiation Therapy (IMRT) and Volumetric Modulated Arc Therapy (VMAT) planning.
• MP demonstrated comparable or superior performance to manual VMAT plans across all evaluated metrics for both IMRT and VMAT.

Conclusions:

• MetaPlanner (MP) offers a generalizable framework for fully automated, high-quality radiation therapy treatment planning.
• The MP method has the potential to substantially reduce the workload for radiation therapy planners.
• This automation can be achieved while maintaining or enhancing the quality of treatment plans.