Cross-engine transformation-based fast dose calculation for MRI-Linac online treatment planning

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Imaging

Background:

• Accurate dose calculation is critical for online treatment planning in Magnetic Resonance Imaging-Linear Accelerators (MRI-Linac).
• Traditional methods may not efficiently handle the complex interplay of magnetic fields and tissue heterogeneity.
• Developing fast and accurate dose calculation algorithms is essential for real-time treatment adjustments.

Purpose of the Study:

• To introduce a novel dose calculation method for MRI-Linac systems.
• The method transforms pencil beam (PB) dose distributions into Monte Carlo (MC) equivalent distributions.
• This aims to enable rapid and accurate online treatment planning.

Main Methods:

• A novel algorithm combines a PB dose engine with a deep learning network (Res-UNet) for magnetic field and tissue inhomogeneity correction.
• The network inputs 3D PB dose data and electron density maps to output corrected dose distributions.
• Trained and validated on 110 IMRT plans across various tumor sites, with performance compared against MC simulations.

Main Results:

• The developed pipeline demonstrated high accuracy, achieving an average plan mean absolute error of 0.90% ± 0.13% and a 98.33% ± 1.07% gamma passing rate (2%/2 mm).
• Excellent consistency was observed between the transformed dose distributions and MC results across all tested sites.
• The entire calculation and transformation process is completed within seconds, enabling rapid online planning.

Conclusions:

• A fast and novel dose calculation pipeline was successfully developed for MRI-Linac online treatment planning.
• The method effectively transforms PB dose distributions to MC-equivalent distributions, accounting for magnetic fields and tissue heterogeneity.
• This approach significantly enhances the feasibility of real-time dose calculations in MRI-guided radiotherapy.