Evaluation of 4‐Hz log files and secondary Monte Carlo dose calculation as patient‐specific quality assurance for VMAT prostate plans

Abstract Purpose In this study, 4‐Hz log files were evaluated with an independent secondary Monte Carlo dose calculation algorithm to reduce the workload for patient‐specific quality assurance (QA) in clinical routine. Materials and Methods A total of 30 randomly selected clinical prostate VMAT plans were included. The used treatment planning system (TPS) was Monaco (Elekta, Crawley), and the secondary dose calculation software was SciMoCa (Scientific‐RT, Munich). Monaco and SciMoCa work with a Monte Carlo algorithm. A plausibility check of Monaco and SciMoCa was performed using an ionization chamber in the BodyPhantom (BP). First, the original Monaco RT plans were verified with SciMoCa (pretreatment QA). Second, the corresponding 4‐Hz log files were converted into RT log file plans and sent to SciMoCa as on‐treatment QA. MLC shift errors were introduced for one prostate plan to determine the sensitivity of on‐treatment QA. For pretreatment and on‐treatment QA, a gamma analysis (2%/1mm/20%) was performed and dosimetric values of PTV and OARs were ascertained in SciMoCa. Results Plausibility check of TPS Monaco vs. BP measurement and SciMoCa vs. BP measurement showed valid accuracy for clinical VMAT QA. Using SciMoCa, there was no significant difference in PTV Dmean between RT plan and RT log file plan. Between pretreatment and on‐treatment QA, PTV metrics, femur right and left showed no significant dosimetric differences as opposed to OARs rectum and bladder. The overall gamma passing rate (GPR) ranged from 96.10% to 100% in pretreatment QA and from 93.50% to 99.80% in on‐treatment QA. MLC shift errors were identified for deviations larger than −0.50 mm and +0.75 mm using overall gamma criterion and PTV Dmean. Conclusion SciMoCa calculations of Monaco RT plans and RT log file plans are in excellent agreement to each other. Therefore, 4‐Hz log files and SciMoCa can replace labor‐intensive phantom‐based measurements as patient‐specific QA.


| INTRODUCTION
In modern radiotherapy techniques, such as volumetric-modulated arc therapy (VMAT), the gantry of the linear accelerator (linac) continuously rotates around the patient while delivering dose. 1,2 VMAT has many advantages compared to conventional radiation techniques, such as improved coverage of the target planning volume (PTV) and minimizing the dose at organs at risks (OARs). 3 For VMAT plans, many different parameters of the linac have to be simultaneously and precisely coordinated to each other (e.g., MLC leaf positions, monitor units [MU], dose rate, gantry angle, collimator angle, jaw position, and energy). Due to the complexity of this radiation technique, consistent quality assurance (QA) for treatment delivery is required. Patient QA guarantees that the patient's correct dosimetry and safety is given. This includes verifying transfer from the treatment planning system (TPS) to the treatment machine and delivery of the calculated dose from the TPS. 4 In general, QA systems can be classified as measurement-based methods (e.g., ionization chamber, 2D and 3D arrays, radiochromic films, EPID) or as simulation-based methods (e.g., log file analysis).
Numerous studies investigate the sensitivity of phantom-based patient-specific QA with induced errors. [5][6][7][8][9][10][11][12][13][14] Other studies demonstrate that measurement devices are inferior to simulation-based systems in detecting errors. [15][16][17] Moreover, using log files for QA makes it possible to automatically check all deliveries in time without the need of a physical phantom and does not require access to the treatment machine. Log file analysis is a very time-saving tool for patient QA in clinical routine. Because of these advantages, there is a growing interest in using log files with an independent Monte Carlo (MC) algorithm. 18 Haga et al. investigated log file analysis for a single simple prostate plan and demonstrated that Elekta 4-Hz log files have satisfying results for QA. 19 Furthermore, Katsuta et al.
showed a strong correlation between log file dose (using MC) and ionization chamber dose (physical measurement). 20 Sun et al. could prove that an independent dose calculation algorithm (convolution superposition algorithm) with log file analysis is a reliable tool for IMRT (intensity-modulated arc therapy) QA. 21 In general, before and during treatment, log file analysis enables us to detect errors that occur when plan data are transferred from the TPS to the linac as well as dose calculation errors or beam delivery errors of the treatment machine. However, research has raised concern about the safety of only using log file analysis for patient QA without any other conventional measurements. 16,22 Hence, it is necessary to investigate accuracy and limits of log file analysis, for example, by comparing measurement-based methods with log file analysis. 17 Ce Han et al. performed a cross verification between measurement-based ArcCHECK QA and simulation-based log file QA using Elekta log files, and it was concluded that sensitivity for log file QA using a collapsed cone (cc) calculation algorithm was superior. 23 In our study, an MC algorithm instead of a cc calculation algorithm was applied to use log files.
Several studies have already dealt with high resolution log files using Varian linear accelerators (Varian Medical Systems, Palo Alto). [24][25][26][27][28][29] Wei Luo et al. demonstrated that Monte Carlo simulation using Dynalog log files (recorded every 50 ms, 20 Hz) has numerous advantages in patient-specific QA compared to measurement-based QA. 24 In measurement-based QA, measurement uncertainties (using film or ionization chamber) might occur, only pretreatment QA is feasible and the information of the measurement in a phantom does not allow drawing conclusions about the dosimetric impact in the patient. However, using Dynalog log files with a Monte Carlo algorithm enables verifying leaf sequencing, data transfer, and beam delivery. Schreibmann et al. also showed that Dynalog log files are a convenient and practical way for dose reconstruction without the need of phantom measurements or phantom calculations. 25   In our study, the investigation of 4-Hz log files was added to further enable on-treatment delivery QA.
Our idea is combining all findings of previously mentioned studies for simple, non-labor-intensive and time-saving patient-specific QA. Log files with a gold standard independent secondary dose calculation algorithm are used to find any kind of errors, including dose calculation errors, transfer errors, and delivery errors. The secondary MC algorithm is first used for plausibility check (TPS dose data) and after irradiation for log file check. This check includes data transfer to the linac and delivery of the linac.
The aim of this study was to evaluate 4-Hz log files with an independent secondary MC dose calculation algorithm to reduce the workload for patient-specific quality assurance in clinical routine to guarantee patient safety. Combining and evaluating an independent secondary MC dose algorithm with 4-Hz log files (Elekta) has, to the authors' knowledge, not been done before.

2.A | Plan selection and treatment planning
A total of 30 randomly selected clinical prostate VMAT plans were included in this study. All plans were calculated with Monaco 5.11.02 (Elekta, Crawley) using Monte Carlo algorithm.
The energy was set to 6 MV. Treatment plans were created using one beam and dual-arc VMAT technique with a fraction of 2 Gy.
The gantry range was −180°to +180°. The collimator angle was fixed during radiation between 0°and 45°depending on the plan.   The treatment delivery of an Elekta linac is controlled by the treatment control system (TCS). It dynamically adjusts the linac parameters to deliver a treatment plan. The TCS works at a frequency of 25 Hz and creates a log file, which is not directly accessible. The linac provides the same data at a rate of 4-Hz at the iCom Interface, which is accessed by LINACwatch® (Qualiformed, France). Kowatsch et al. showed that the difference between 25-Hz and 4-Hz log files is negligible for dose calculations. 34 All log files contain several dynamic parameters such as leaf position, gantry position, collimator position and delivered monitor units. For reading and converting log files, the software LINACwatch was used. All log files had to be converted into DICOM RT plan files and were then referred to as RT log file plans, which were generated with as many control points as in the log file.

2.E | Pretreatment and on-treatment patientspecific QA
Patient-specific QA was subdivided into pretreatment and ontreatment QA. Both methods are fundamental for reliable patientspecific QA to guarentee treatment safety of the patient. Patientspecific QA is described in Fig. 1

2.F | Sensitivity of MLC errors
To verify that error induced plans can be detected as such, one prostate plan was manipulated with an in-house Matlab® (Math-Works Inc., Natick) tool. MLC misalignments with different magnitudes were applied (MLC opening and MLC closing from 0.25 to 0.75 mm, increment 0.25 mm). All six error induced plans were delivered by the linac and the corresponding log file was compared with the reference plan from TPS. For all error induced plans an ontreatment QA was performed.
A gamma criterion of 2%/1 mm/20% (threshold) and a pass limit of 90% for pretreatment and on-treatment QA for overall gamma was used. Moreover, PTV Dmean was set to 2% dose tolerance and a visual comparison of PTV coverage and OARs in DVH was performed by an experienced physicist.

2.G | Commissioning and validation of Monaco and
SciMoCa TPS Monaco beam data were collected applying Elekta guidelines and validated in 2013. The commissioning process for the XVMC dose algorithm was done by Elekta. Implementing MC-based systems into clinical routine is well described in literature. 35,36 The introduced software SciMoCa uses a more precise Monte Carlo algorithm, which has already been reported in literature. 31

2.H | Statistical analysis
Statistical analyses were performed using SPSS 23 (IBM, New York).
Mean dose results and GPRs are presented as mean ± 1 SD. A p value smaller than 0.05 was defined as statistically significant and all p values are two-sided. The analysis of correlation coefficient r was done according to Pearson. The Pearson correlation coefficient r was F I G . 1. Flow chart of the entire pretreatment and on-treatment QA used in this study. Pretreatment QA uses the RT plan of TPS Monaco and compares its dose distribution calculated in Monaco vs. SciMoCa using original CT images of the patient. In on-treatment QA, the RT plan is delivered by the linac and then converted into an RT log file plan by the software LINACwatch. On-treatment compares the TPS dose distribution of the RT plan with the dose distribution of the RT log file plan calculated with SciMoCa.  Target planning volume Dmean was compared between RT plan and RT log file plan in SciMoCa, which can be seen in Fig. 4. For RT plan calculation, the PTV Dmean was 2.03 ± 0.01 Gy and for RT log file plan 2.03 ± 0.02 Gy (p = 0.21). There was a strong correlation for PTV Dmean between RT plan and RT log file plan with r = 0.97 (p < 0.001).

3.B | Pretreatment and on-treatment patientspecific QA
Detailed dosimetric differences for different metrics for all prostate plans in pretreatment and on-treatment patient-specific QA are shown in Table 1. In all PTV metrics, in femur right and in femur left, no significant dosimetric differences were observed between pretreatment and on-treatment QA. For the OARs rectum and bladder significant dosimetric differences occurred between pretreatment and on-treatment QA.

3.C | Sensitivity of MLC errors
One prostate plan was manipulated to see if the described pretreatment and on-treatment patient-specific QA programs can identify these errors.  Table 3.
On-treatment QA for error induced prostate plans identified MLC shift errors for deviations larger than −0.50 mm and +0.75 mm with the limits set to 2%/1 mm/20% (threshold) for overall gamma criterion, a pass limit of 90% and 2% for PTV Dmean difference (ΔPTV Dmean). Further details for MLC shift error are extracted from dose distribution and DVH analysis in SciMoCa. Figure 7 shows the dose distribution and DVH analysis of on-treatment QA for MLC shift error of −0.75 mm.

| DISCUSSION
Using 4-Hz log files with an independent secondary Monte Carlo dose calculation algorithm enables on-treatment QA for every fraction. When using log files for patient-specific QA, discrepancies between the original Monaco RT plan, which drives the linac, and the RT log file plan generated by the linac are not acceptable.
Thus, calculations with both RT plans (Monaco RT plan and RT log file plan) must lead to the same results applying the same calculation method. This is a fundamental requirement to use log files in patient-specific QA. In this study, all PTV metrics (see Fig. 4 and Table 1) yielded the same results for RT plan and RT log file plan with 4-Hz log files using SciMoCa. Furthermore, the GPR calculations for pretreatment and on-treatment patientspecific QA are consistent (see Table 2). If there were disagree-   However, the sensitivity of their measurement-based QA was inferior to the performance of our on-treatment QA using log files.
A significant advantage of patient-specific QA with log files is that the influence of MLC shift errors on PTV Dmean can directly be seen in the patient's dose distribution in SciMoCa. In measurement-based QA, PTV Dmean cannot be assessed.
In comparison to phantom-based QA measurements, using log files is very time-saving (calculation time of just about 5 min per fraction using MC) and much more efficient in finding errors. 15,16 Furthermore, log file analysis with simple fluence calculation outperforms ArcCHECK (3D Array, Sun Nuclear) measurements due to sensitivity for VMAT plans. 17 The advantage of using an independent In general, all recorded log files are insensitive to miscalibration. 16,38,39 Accuracy of log files is crucial for patient-specific QA.

MLC shift error
On-treatment QA There are limitations to this study, since only prostate plans were analyzed. Moreover, every calculation was performed with the planning CT instead of the daily CT at the linac and no changes in anatomy or patient position were taken into consideration. 41 The next step is to include the daily CT into our described on-treatment patient-specific QA program to see dose distribution of the day in the actual anatomy of the patient.

| CONCLUSION
Our study demonstrates that 4-Hz log files and an independent secondary MC dose calculation algorithm have the potential to replace