TL;DR
This paper tackles the challenge of creating a single, unified treatment planning system (TPS) model for multiple identical linear accelerators (linacs) in a clinical setting. The authors successfully developed and validated such a model in RayStation for four Elekta VersaHD linacs, demonstrating its efficiency and effectiveness compared to individual linac models.
What Problem Does This Paper Address?
Modern radiation therapy relies on sophisticated treatment planning systems (TPS) to deliver precise radiation doses to patients. When a clinic operates multiple linear accelerators (linacs), especially identical ones, it's highly beneficial to have a single TPS model that accurately represents all of them. This "single model" approach streamlines workflows, allows for flexible patient assignments, simplifies quality assurance (QA), and provides a backup solution in case of machine downtime. However, building and validating such a unified model, particularly for newer TPS like RayStation, isn't always straightforward, and established guidelines are still evolving. This paper addresses the need for a practical and efficient strategy to commission and validate a single photon beam model in RayStation for a group of matched Elekta linacs.
Methods
The authors employed a multi-step approach to build and validate their RayStation photon beam model.
- Data Compilation and Comparison: Beam data from all four Elekta VersaHD linacs were collected and compared to identify median values for model parameterization.
- Automated Modeling: RayStation's built-in multi-parameter and single-parameter auto-modeling tools were utilized to generate and refine the initial beam model, leveraging the collected data. Manual adjustments were made to optimize parameters like MLC and jaw positions.
- Validation Strategy: Recognizing that full validation on all four linacs would be time-consuming, they adopted a hybrid approach. A comprehensive MPPG-5a validation was performed on one linac, followed by spot checks on the other three. This was supplemented by simple static field validation and an extensive IMRT/VMAT QA validation using over 100 patient cases. An external end-to-end test using IROC anthropomorphic phantoms was also conducted for each linac.
- Curve Quality Metric: The Root Mean Square (RMS) difference was used to quantify the agreement between measured and computed data for PDD and profiles, with specific tolerance criteria applied.
Datasets
The study utilized existing clinical data and established QA procedures.
- Beam Data: Measured beam data (PDD, profiles, output factors, wedge factors) from four Elekta VersaHD linacs commissioned between 2013 and 2020.
- Patient-Specific IMRT/VMAT QA Data: 103 previously treated patient cases (planned in Pinnacle v16.2 and QA'd with MapCHECK2) were re-analyzed using the new RayStation model. Additionally, QA results from 228 6-MV, 120 6-MV-FFF, and 45 10-MV cases treated with RayStation after its clinical implementation were collected.
- IROC Anthropomorphic Phantoms: Four phantoms (Head and Neck, Lung, Spine, Prostate) were used for external end-to-end testing, with embedded films and TLDs for analysis.
Key Insights and Ideas
- Efficiency of Automated Tools: RayStation's auto-modeling tools significantly expedited the beam modeling process, reducing the need for extensive manual data acquisition and parameter tuning.
- Hybrid Validation is Effective: For closely matched linacs, a comprehensive validation on one machine, coupled with targeted spot checks and extensive IMRT/VMAT QA on others, is a time-efficient and robust strategy.
- Single Model Performance: The single RayStation model performed comparably to, and in some aspects better than, the previously used individual linac models in Pinnacle, particularly in IMRT QA passing rates.
- Practical Approach for Institutions: The proposed method offers a practical blueprint for other institutions looking to implement a single TPS model for multiple matched linacs, especially when transitioning between TPS vendors.
Key Results
- PDD Agreement: The single model's PDDs were within 1% agreement of the four-linac commissioning measurements across all energies.
- MPPG-5a and TG-119 Compliance: All MPPG-5a validation tests and TG-119 measurements passed within the recommended tolerance limits.
- IMRT QA Comparison: The RayStation single model achieved high IMRT QA passing rates, with mean ± standard deviation values such as 98.9% ± 1.3% for 6-MV (compared to 98.0% ± 1.4% for individual Pinnacle models).
- External End-to-End Test: Results from the IROC phantom tests demonstrated the accuracy and reliability of the RayStation model across different linacs and treatment sites.
- Speed Advantage: The paper notes that RayStation's GPU server-driven dose computation speed outpaces Pinnacle by 1-2 orders of magnitude.
Limitations and Future Work
- Kernel Tilting: RayStation's dose engine lacks kernel tilting, which can lead to more pronounced errors in profiles for larger fields and greater depths, particularly in the penumbra area. This is noted as an area for improvement, though less critical for IMRT.
- Wedge Model Fidelity: The RayStation wedge model is tied to the open field model of the corresponding energy, offering less fine-tuning flexibility compared to Pinnacle's independent wedge model. The authors suggest an independent approach in RayStation could improve wedge field fidelity.
- High Z Material Discrepancies: Discrepancies were observed in dose calculations near high Z materials due to RayStation's photon beam minimum energy bin being set at 0.5 MeV, compared to Pinnacle's 0.25 MeV. This resulted in artificially smoother dose profiles in RayStation for certain scenarios.
- Future Work: The authors advocate for adopting a similar independent approach for wedge models in RayStation to enhance fidelity. They also suggest that their proposed validation methods can serve as a guide for other institutions.
Summary updated