An Independent Three-Dimensional (3D) Dose Veri�cation System for Elekta Unity MR-Linac Online Plans

Purpose To implement an independent 3D dose veri�cation system with RayStation (RaySearch, Stockholm, Sweden) for online adaptive radiotherapy on Elekta Unity MR-Linac (MRL). Methods Plan quality of simple-single-eld and intensity-modulated radiotherapy (IMRT) plans were investigated in a comparison of (1) Monte-Carlo calculated data using MRL Monaco with high magnetic �eld (1.5 T) and (2) Collapsed-Cone calculated data using RayStation. The dose quality of RayStation plans, compared to corresponding Monaco plans, was (1) visually inspected in percentage depth-dose curves, inline and crossline pro�les, and (2) quanti�ed in 3D gamma-passing-rates. Processing time was measured to evaluate the practical e�cacy of our system using 5 prostate IMRT plans.


Introduction
Patient speci c pre-treatment quality assurance (QA) is essential to ensure the plan quality of individual treatment plans. 1,24][5] However, these QA devices are insu cient for online adaptive radiotherapy due to that it requires online QA(s) during day-to-day plan adaptations. 6To overcome this issue, an independent dose veri cation system is required to con rm plan quality without interrupting online adaptive radiotherapy.
To perform dose veri cation for an Elekta Unity (Elekta AB, Stockholm, Sweden) MRI-Linac (MRL) equipped with a 1.5 T MRI scanner (Philips Healthcare) during online adaptive radiotherapy, Chen, et al., developed an in-house software tool to account for the in uence of the high magnetic eld. 7They considered transverse magnetic eld across several depth doses and various pro les in dose calculation.
In addition, a commercial software package (RadCalc v6.3, LifeLine Software, Inc.) was evaluated for the Elekta Unity MRI-Linac (Elekta AB, Stockholm, Sweden) to verify plan quality prior to patient treatment. 8am pro les affected by a 1.5 T magnetic eld at gantry 0 o and 270 o for the eld sizes of various simple single elds (i.e., square and rectangular shapes) at several depths were utilized to build a relative beam model on the RadCalc.
The third party treatment planning systems (TPS) were utilized to build individual MRL beam models using asymmetry dose pro les calculated on a Monte Carlo algorithm on Monaco v5.4 (Elekta AB, Stockholm, Sweden) due to the effect of a 1.5 T magnetic eld.Goodwin, et al. 9 and Li, et al., 10 were utilized RayStation TPS(s) to build a MRL beam model and commission the MRL beam model for developing RayStation plans.Then, the plan quality of MRL Monaco plans, compared to corresponding RayStation plans was veri ed by visually inspecting dose difference 9 and quantifying 3D gamma pass rates at an average 90% in 3%/3mm gamma criteria. 10In addition, the Pinnacle (Philips, Best, the Netherlands) TPS was used to develop quasi MRL plans incorporated MRL characteristics and compare with MRL plans developed on Monaco v5. 4. 11 In overall, online adaptive radiotherapy requires an independent dose calculation system for the secondary dose veri cation on MRL.However, it has dependency on the characteristics of individual MRL(s) and TPS(s), and commercial software currently provides a comparison of a single point dose 8 and the third party TPS also provides an average 90% of plan quality in 3D doses.Hence, this study aims to implement an independent and 3D dose veri cation system with RayStation for online QA during online adaptive radiotherapy planning.We devised in-house gamma analysis software, which can compare two plans, and provides 3D volumetric gamma analysis.

Methods
An independent 3D dose veri cation system using RayStation (Version 9A) is comprised of ve steps: (1)   developing MRL plans on an MRL Monaco (Version 5.40.01), (2) exporting Monaco DICOM data to the pre-determined folder, (3) importing the Monaco DICOM data to RayStation and dose calculation, (4)   exporting RayStation DICOM data (only plan and dose) to the same pre-determined folder, and (5)   comparing Monaco and RayStation plan doses using an in-house 3D gamma analysis software.Figure 1 shows the work ow of an independent 3D dose veri cation system in ve steps.

Developing MRL Monaco plans
Various simple single eld plans (i.e., simple 3D plans with a single square and rectangular eld) were made using MRL Monaco with Monte Carlo® algorithm.Those plans were calculated using GPUMCD (GPU-oriented Monte Carlo dose calculation algorithm) in MRL Monaco TPS with high magnetic eld (1.5 T) and a 7MV FFF ( attening lter free) energy.The simple single eld plans with 100 monitor units (MU) at SSD 133.5 cm across various eld sizes (3×3 cm 2 , 5×5 cm 2 , 10×10 cm 2 , 10×10 cm 2 , 15×15 cm 2 , 20×20 cm 2 , 22×22 cm 2 and 40×22 cm 2 ) and gantry angles (0 o , 90 o , 180 o and 270 o ) were developed using a uniform water phantom and an MRL couch.The MR coil was absent during the development of simple single eld plans.A 0.3 cm grid spacing and a 0.5% statistical uncertainty were used during dose calculation for all simple single eld plans.These Monaco simple single eld plans were used as inputs of a beam model built on RayStation and following the beam model veri cation whilst comparing to RayStation simple single eld plans (see the Sect.2.2).
In addition, 7-eld intensity-modulated radiation therapy (IMRT) plans were developed on MRL Monaco using volumetric modulated arc therapy (VMAT) plans of ve prostate cancer patients received external beam radiotherapy.The CT images and structures of VMAT plans were manually exported from Monaco v5.11.02 (Elekta AB, Stockholm, Sweden) as a DICOM format and imported to MRL Monaco.The same CT images, structures and prescriptions (7000 cGy with 28 fractions for a plan and 4600 cGy with 23 fractions for 4 plans) were utilized during the plan development of 7-eld IMRT plans.For the 7-eld IMRT plans, a MV 7FFF beam was used with MRL plan parameters (i.e., Gantry angles at every 51 o , collimator 0 o , the isocenter location at the center of planning target volume, dose rates at 425 MU/min, and MR coil and MR couch components).A 0.3 cm grid spacing and a 0.5% statistical uncertainty were used during dose calculation for all 7-eld IMRT plans.

Developing MRL RayStation plans
Prior to developing MRL RayStation plans, an MRL beam model was built on RayStation using inline and crossline pro les, and percentage depth dose (PDD) curves of the Monaco simple single eld plans which have a zero gantry angle.The PDD curves and pro les were extracted from multiple depth doses at maximum dose, 5 cm, 10 cm and 20 cm as inputs utilized during RayStation beam modeling.Pro les and offsets of an MRL RayStation beam were adjusted for best matching to the corresponding the input pro les across all elds and depths.RayStation simple single eld plans were then developed using the new MRL RayStation beam and a Collapsed Cone algorithm to calculate plan doses.The same uniform water phantom and MRL couch were used in the development of Monaco simple single eld plans.The same grid spacing and statistical uncertainty of Monaco plans were used during dose calculation for all simple single eld plans.We intended to isolate the source of discrepancy possibly occurred in beam commissioning process before proceeding to complex IMRT plan comparison.
In addition, a DICOM importing tool was developed using Python scripts in RayStation to automatically modify DICOM headers, accounting for the difference of DICOM properties between Monaco and RayStation.It copied DICOM properties which are absent such as SoftwareVersion and FrameOfReferenceUID from a reference CT image to a radiotherapy (RT) Plan.An essential DICOM properties such as an x MLC component was additionally added to BeamLimitingDeviceSequence due to the absence of the x MLC component in the MRL Monaco plan.During importing MRL Monaco plans, the tool automatically allocated a CT scanner and a treatment machine by matching their identical names to one of the pre-registered CT scanners and treatment machines.
To develop MRL RayStation plans, our scripts allowed streamlining the process, so that the MRL Monaco plan, which was presently imported without the modi cation of plan parameters, was utilized to calculate plan dose using the Collapsed Cone algorithm.The same grid spacing and statistical uncertainty of Monaco plans was used during dose calculation for all 7-eld IMRT plans.The grid spacing was manually set in this study after completing our scripts.

Dose comparisons and statistical analysis
RayStation simple single eld plans were compared to corresponding Monaco simple single eld plans.
For dose comparisons between RayStation and Monaco simple single eld plans, our in-house gamma analysis software imported and used plan and dose les, and analyzed 3D gamma passing rates. 12amma analysis with 3%/3mm and 2%/2mm was performed and 3D gamma passing rates were quanti ed to compare dose similarity between RayStation and Monaco plans.The gamma passing rates of the simple single eld plans were compared to assess the dependency of gantry angles at 0 o , 90 o , 180 o and 270 o .Similarly, RayStation 7-eld IMRT plans were compared to corresponding Monaco 7-eld IMRT plans.For dose comparisons between RayStation and Monaco 7-eld IMRT plans, our in-house gamma analysis software was used to perform gamma analysis with 3%/3mm gamma criteria.
Using Monaco 7-eld IMRT plans, the performance of the proposed independent 3D dose veri cation system was assessed by measuring the processing time of (1) exporting DICOM data from MRL Monaco, (2) importing DICOM data to RayStation via scripting, (3) calculating plan dose using Collapsed Cone algorithm, (4) exporting DICOM data from RayStation and (5) comparing plan dose using the in-house 3D gamma analysis software.

Results
In this study, an independent 3D dose veri cation system was successfully developed by (1) building a new MRL beam model for developing simple single eld plans and 7-eld IMRT plans on RayStation, (2) implementing Python scripts for importing MRL Monaco plans to RayStation and (3) implementing an inhouse software for gamma analysis.

Dose comparison between simple single eld plans
MRL Monaco simple single eld plans were compared to corresponding MRL RayStation simple single eld plans by using PDD curves and dose pro les, and quantifying 3D dose differences in gamma analysis.A PDD curve, inline (superior to inferior) and crossline (left to right) dose pro les extracted from each 3D dose at the central axis and they were visually inspected.Figure 2 shows an example of a visual inspection from a simple single eld plan with a 10 × 10 cm 2 eld size and a 10 cm depth.Both doses of MRL Monaco simple single eld plans using Monte Carlo and RayStation simple single eld plans using collapsed cone algorithm were normalized to 100 cGy at a 10 cm depth.
Both simple single eld plans were well agreed in the PDD curve, inline and crossline pro les.The maximum dose of both Monaco and RayStation simple single eld plans in the PDD curve (see Fig. 2(a)) was about 146 cGy at a 1.5 cm depth.Inline and crossline pro les (see Fig. 2(b) and (c)) of both Monaco and RayStation simple single eld plans were very similar.
The quality of simple single eld plans measured by comparing 3D doses using the in-house gamma analysis software is shown in Fig. 3.
The RayStation simple single eld plans with the new MRL beam model showed promising results with an average 95.7% and 98.5% in 2%/2mm and 3%/3mm gamma criteria (see Fig. 3(a) and (b)), respectively.With the same gantry angle, the eld size dependency of RayStation simple single eld plans in 2%/2mm gamma criteria was 3.6% at 0 o , 4.6% at 90 o , 4.0% at 180 o and 6.0% at 270 o but it was < 3.2% for all gantry angles when excluded the 3cm × 3cm eld.In 3%/3mm gamma criteria, it was ≤ 1.9% for all eld sizes.
With the same eld size, the gantry angle dependency of simple single eld plans in 2%/2mm gamma criteria was 2.2% at 3cm × 3cm, 2.5% at 5cm × 5cm, 2.8% at 10cm × 10cm, 3.5% at 20cm × 20cm and 3.0% at 22cm × 22cm.In 3%/3mm gamma criteria, it was < 2.2% for all gantry angles.The gamma passing rate of the same eld size (see Fig. 3(b)) was the smallest at gantry angle 180 o due to the attenuation of a posterior MR receiver coil (i.e., about 0.9% attenuation) and it was very similar at both gantry angle 90 o and 270 o .

Dose comparison between 7-eld IMRT plans
MRL Monaco 7-eld IMRT plans were compared to corresponding MRL RayStation 7-eld IMRT plans by using dose pro les and quantifying 3D dose differences in gamma analysis.Inline and crossline dose pro les extracted from each 3D dose at the central axis and they were visually inspected.Figure 4 shows an example of a visual inspection from a 7-eld IMRT plan.
Both Monaco and RayStation 7-eld IMRT plans were reasonably well agreed in the longitudinal, inline and crossline pro les.The doses of the Monaco and RayStation plans at the central axis were 4.4% higher (261 cGy) and − 1.3% lower (247 cGy) from 250 cGy, respectively.Figure 5 shows an example of the quality of 7-eld IMRT plans (7000 cGy in 28 fractions) which comparing 3D doses using the in-house gamma analysis software.The Monaco 7-eld IMRT plan and dose were displayed on RayStation to directly compare to RayStation 7-eld IMRT plan and dose.
Both Monaco and RayStation 7-eld IMRT plan doses are visually very similar and the target (a solid blue line in Fig. 5(a) and (b)) entirely remains in yellow and orange colors in the range between 7271 cGy (95% of 7654 cGy) and 6889 cGy (90% of 7654 cGy), respectively.Their dose difference in gamma remains in less than 3.0% which the red color (see Fig. 5(c)) was minimal except for the entry of beams.

Performance assessment of the independent dose veri cation system
An independent 3D dose veri cation system using Monaco 7-eld IMRT plans was assessed by measuring the processing time of individual steps as a function of its performance.An average of the total processing time was approximately 200 s which summed individual processing time of (1) DICOM export from MR-Linac Monaco (34 s), (2) DICOM import to RayStation via scripting (80 s), (3) dose calculation (24 s), (4) DICOM export from RayStation (26 s), and (5) 3D Gamma analysis using the inhouse gamma analysis software (36 s).

Discussion
Online adaptive radiotherapy using MRL requires an independent dose veri cation to con rm plan quality in a short period time in the presence of patients on the treatment couch.This study implemented an independent 3D dose veri cation system using RayStation and in-house 3D gamma analysis software.Using the independent 3D dose veri cation system, we demonstrated (1) a superior plan quality (< 98.5% in 3%/3mm gamma criteria) of RayStation simple single eld and 7-eld IMRT plans whilst comparing to corresponding Monaco simple single eld and 7-eld IMRT plans, and (2) its applicability for independently verifying plan quality during MRL online adaptive radiotherapy.
An independent 3D dose veri cation needs to account for a 1.5 T high magnetic eld 7− 9 so this study also utilized Monte Carlo® calculated data (i.e., Asymmetry pro le data) during beam modeling on RayStation.In the use of the beam model, RayStation simple single eld plans compared to corresponding Monaco simple single eld plans achieved an average 95.7% and 98.7% of plan quality in 2%/2mm and 3%/3mm gamma criteria (see Fig. 3), respectively.In addition, RayStation 7-eld IMRT plans achieved an average 95.1% in 3%/3mm gamma criteria, which can be (1) acceptable to minimize dependency of individual MRL(s) and ( 2) applicable to verify plan quality during the online QA(s) of dayto-day plan adaptations. 6eed can be essential to minimize the variability of organ locations and changing the position of patient setups during dose veri cation and also e ciency can be very important to improve the performance of an independent 3D dose veri cation system.This study implemented Python scripts for importing Monaco DICOM data to RayStation and an in-house 3D gamma analysis software for comparing plan quality, resulted in that entire dose veri cation can be achievable in approximately 200 s.In addition, automated processes reduced speed and improve e ciency whilst minimizing the involvement of manual processes (i.e., only DICOM export to a pre-determined folder).
One of the limitations of the present study was the dependency of gantry angles (see Fig. 3).The accuracy of dose calculation can be determined by precise beam modeling on individual TPS(s) [7][8][9][10][11] and thus our beam model was built on RayStation using the characteristic and speci c information of MRL. 13,14However, it still included about 2.0% of gantry angle dependency (see Fig. 3(b)) in 3%/3mm gamma criteria, requiring a compensation of gantry angle dependency. 9 implemented the in-house software only for comparing 3D dose to analyze plan quality in gamma criteria and also this study did not consider more complicate plans.For better and powerful analysis of plan quality, the dose volume histogram and dose coverage of target and organ at risks could be utilized during online QA and reviewing adaptive plans prior to daily treatment delivery.

Conclusion
This was the study to implement an independent 3D dose veri cation system using RayStation with scripts and an in-house 3D gamma analysis software.This led to an average 95% of plan quality in 3%/3mm gamma criteria and added an average 200 s throughout the entire veri cation process in the independent 3D dose veri cation system.These results demonstrate that this approach can be applicable and e cient for online QA during online adaptive radiotherapy with minimal time extension.Further investigations will need to include more heterogeneous target and organ at risk samples.

Figures
Figure 1 The work ow of an independent 3D dose veri cation system.(a) Developing MRL Monaco plans, (b) exporting Monaco DICOM data to a pre-determined folder, (c) importing the Monaco DICOM data to RayStation and dose calculation, (d) exporting RayStation DICOM data (only plan and dose) to the same pre-determined folder, and (d) comparing Monaco and RayStation plan doses using an in-house 3D gamma analysis software.

Figure 2 An
Figure 2

Figure 3 The
Figure 3

Figure 4 An
Figure 4