Commissioning dose computation model for proton source in pencil beam scanning therapy by convolution neural networks
Yaoying Liu1,2,3, Xuying Shang1,2,3,4, Wei Zhao1,5
1School of Physics, Beihang University, Beijing, 102206, People's Republic of China.
Physics in Medicine and Biology
|July 5, 2023
Summary
We developed PSMC-Net, a novel deep learning method for proton source model commissioning. This approach significantly improves the accuracy of Monte Carlo simulations in proton therapy dose calculations.
Area of Science:
- Medical Physics
- Radiation Oncology
- Computational Biology
Background:
- Proton source model commissioning (PSMC) is essential for accurate dose calculations in pencil beam scanning (PBS) proton therapy using Monte Carlo (MC) simulations.
- Commissioning nominal energy and energy spread parameters in PSMC is challenging due to their indirect nature.
Purpose of the Study:
- To develop an efficient and accurate method for commissioning proton source model parameters.
- To introduce a novel convolution neural network (CNN) named 'PSMC-Net' for PSMC.
Main Methods:
- PSMC-Net was trained on datasets of source model parameters and corresponding MC integrated depth doses (IDDs).
- Separate training was performed for 33 energies between 70-225 MeV.
- The network was trained using 130 data pairs for training, 10 for validation, and 10 for testing.
Main Results:
- The commissioned source model using PSMC-Net achieved a gamma passing rate (GPR) of 99.91 ± 0.12% (1 mm/1%) compared to measured IDDs.
- Without commissioning, the GPR dropped significantly to 54.11 ± 22.36%.
- Average 3D GPRs for clinical cases (2 mm/2%) were 99.89% for a spread-out Bragg peak and 99.96 ± 0.06% for patient plans.
Conclusions:
- PSMC-Net offers a novel, easy, efficient, and accurate method for proton source model commissioning.
- This deep learning approach significantly enhances the precision of MC-based dose calculations in proton therapy.
