A time- and space-saving Monte Carlo simulation method using post-collimation generative adversarial network for dose...

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Computational Modeling

Background:

• The RefleXion X1 Linac is an advanced radiotherapy machine featuring a binary multi-leaf collimation (MLC) system.
• This system enables innovative biology-guided radiotherapy techniques.

Purpose of the Study:

• To assess the feasibility of using Generative Adversarial Networks (GANs) for modeling the RefleXion X1 Linac.
• To evaluate the accuracy of GAN-based dose simulations.
• To determine the potential computational benefits of this approach.

Main Methods:

• Developed 34 GAN generators, each trained on phase space files for specific MLC apertures.
• Simulated dose distributions in water using both GAN and phase space sources for comparison.
• Estimated computational time savings and storage reduction by bypassing collimation simulation.

Main Results:

• GAN simulations showed high agreement with phase space simulations for percentage depth dose, penumbra, and full-width half maximum.
• Gamma passing rates (1%/1mm) exceeded 90% for all apertures.
• Achieved an estimated time saving of 530 CPU hours and a 102-fold reduction in storage usage for a plan with 5766 beamlets.

Conclusions:

• Generative Adversarial Networks (GANs) provide an accurate and efficient method for simulating the RefleXion X1 Linac.
• The significant reductions in computational time and storage make GANs highly valuable for future dosimetry and beam modeling in radiotherapy.