Accelerated robust optimization algorithm for proton therapy treatment planning.

Area of Science:

• Medical Physics
• Computational Optimization
• Radiation Oncology

Background:

• Robust optimization in radiation therapy is computationally intensive, leading to long plan computation times, especially for moving targets requiring numerous uncertainty scenarios.
• Conventional minimax optimization struggles with efficiency due to the extensive number of scenarios evaluated.

Purpose of the Study:

• To introduce a novel worst-case robust optimization algorithm, dynamic minimax, designed to accelerate the conventional minimax optimization process.
• To decrease the number of evaluated scenarios during robust optimization for improved efficiency.

Main Methods:

• The dynamic minimax algorithm selects a reduced subset of candidate-worst scenarios from a larger pool, updating them throughout the optimization using a probability acceptance function.
• Implementation in the open-source robust optimizer MIROpt and testing on six 4D Intensity Modulated Particle Therapy (IMPT) lung cancer patients.
• Robustness testing involved simulating range errors, setup errors, and breathing motion, with dose calculations performed using the MCsquare Monte Carlo dose engine.

Main Results:

• The dynamic minimax algorithm achieved an average optimization time gain of 84%.
• While the optimization process was noisier, the final plan quality (target coverage and normal tissue sparing) remained comparable to conventional minimax optimization.
• Differences in worst-case D95, mean lung dose, and mean heart dose were minimal (0.2 Gy, 0.4 Gy, and 0.1 Gy, respectively).

Conclusions:

• The dynamic minimax algorithm offers a significant 84% optimization time reduction for worst-case 4D robust optimization.
• This acceleration is achieved without any compromise in target coverage or normal tissue sparing, making it a valuable advancement in radiation therapy planning.