Robust deep learning-based forward dose calculations for VMAT on the 1.5T MR-linac
G Tsekas1, G H Bol1, B W Raaymakers1
1Department of Radiotherapy, University Medical Center Utrecht, Heidelberglaan 100, Utrecht 3584CX, The Netherlands.
Physics in Medicine and Biology
|October 5, 2022
Summary
This study introduces a deep learning framework for accurate dose calculations in Volumetric Modulated Arc Therapy (VMAT) on a 1.5T MR-linac. The method ensures robust and precise treatment planning for various cancer types.
Area of Science:
- Medical Physics
- Radiotherapy Technology
- Artificial Intelligence in Medicine
Background:
- Volumetric Modulated Arc Therapy (VMAT) is a widely used radiotherapy technique.
- Accurate dose calculation is crucial for effective VMAT delivery, especially with advanced systems like the 1.5T MR-linac.
- Deep learning offers potential for improving computational efficiency and accuracy in dose prediction.
Purpose of the Study:
- To develop and validate a robust deep learning framework for VMAT forward dose calculations on a 1.5T MR-linac.
- To assess the accuracy and reliability of the proposed method for diverse tumor sites and patient data.
- To establish a promising solution for VMAT plan validation in clinical workflows.
Main Methods:
- A convolutional neural network (CNN) was trained using VMAT segment doses from clinical data converted to an MR-linac environment.
- The CNN predicted segment doses for MR-linac-deliverable VMAT test plans, with randomized angles and shifted MLC shapes for robust learning.
- Monte Carlo simulations generated ground truth dose data with 1% uncertainty.
Main Results:
- The deep learning framework achieved highly accurate dose distributions for MR-linac VMAT plans, with 99.7% ± 0.5% passing the 3%/3 mm gamma criterion.
- Evaluation on unseen Intensity-Modulated Radiation Therapy (IMRT) patients also met clinical requirements, achieving 99.0% ± 0.6% for the 3%/3 mm analysis.
- The method demonstrated robustness across different tumor anatomies and tissue densities.
Conclusions:
- The presented deep learning framework provides a robust and accurate solution for VMAT forward dose calculations on the 1.5T MR-linac.
- This approach shows significant promise as a plan validation tool for both IMRT and VMAT, enhancing treatment planning safety and efficiency.
- The framework's ability to handle diverse patient data and anatomical variations underscores its clinical applicability.
