Automating proton treatment planning with beam angle selection using Bayesian optimization

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Biology

Background:

• Proton therapy offers precise dose delivery, but treatment planning, especially beam angle selection, remains complex and time-consuming.
• Automating the treatment planning process is crucial for improving efficiency and consistency in proton therapy.

Purpose of the Study:

• To develop a fully automated treatment planning process for proton therapy.
• To integrate a novel Bayesian optimization approach with a constrained hierarchical optimization method (ECHO) for automated beam angle selection.

Main Methods:

• Adapted an existing automated intensity modulated radiation therapy (IMRT) planning system (ECHO) for proton therapy.
• Implemented a Bayesian optimization technique for selecting optimal beam angles.
• Evaluated the integrated system on five head-and-neck cancer patients using various beam configurations (coplanar and noncoplanar).

Main Results:

• The Bayesian optimization identified optimal beam configurations in a small fraction of potential options (<4% for coplanar, <1% for noncoplanar).
• Automated planning reduced average mandible maximum dose by 6.6 Gy and dose to unspecified normal tissues by 3.8 Gy compared to planner-selected configurations.
• The algorithm efficiently converged, demonstrating its applicability to complex beam arrangements.

Conclusions:

• A fully automated and efficient proton therapy treatment planning process, including beam angle optimization, has been successfully developed.
• The combination of Bayesian optimization and ECHO generates high-quality treatment plans with optimal beam configurations.
• The framework's ability to handle complex objective functions allows incorporation of various clinically relevant metrics for dose evaluation.