INTRODUCTION
Rapid ArcTM (Varian Medical Systems, Palo Alto, CA, USA) is a commercial method of delivering volumetric-modulated arc therapy (VMAT) treatment plans in one or more arcs with continuously variable gantry speed, MLC positioning and dose-rate during gantry rotation. Therefore, due to the increased degree of modulation and complexity of the VMAT plans in comparison to the traditional intensity-modulated radiation therapy (IMRT) plans, it is necessary to perform a pre-treatment patient-specific quality assurance (QA). The goal of pre-treatment patient-specific QA is to check that the linear accelerator is capable of delivering the dose distribution that was calculated by the treatment planning system (TPS). The most widely used QA procedure is to deliver the plan on a two-dimensional (2D) or three-dimensional (3D) ion chamber or diode arrays to obtain a measured dose distribution.Reference Kim, Park, Kim, Kim, Ye and Park 1 The measured dose distribution is then compared with the dose distribution calculated by the TPS and the gamma index is commonly used to quantitatively compare the two dose distributions.Reference Low, Harms, Mutic and Purdy 2 ArcCHECKTM (Sun Nuclear Corp., Melbourne, FL, USA) and Delta4 TM (ScandiDos AB, Uppsala, Sweden) are two commercial systems that are widely used for patient-specific QA.Reference Liang, Liu, Zhou, Yin and Wu 3 The electronic portal imaging device (EPID) that is integrated with Linacs is another option for patient-specific QA and can be used with no additional cost of purchasing an imager since it is part of the Linac. EPID could potentially be an attractive alternative to the other systems due to its simplicity in setup.
Several studies have been performed to investigate the sensitivity of gamma index to detect simulated potential delivery errors such as in multileaf collimator (MLC), couch, gantry, collimator positioning errors and machine output during VMAT and IMRT treatments.Reference Kim, Park, Kim, Kim, Ye and Park 1 , Reference Liang, Liu, Zhou, Yin and Wu 3 – Reference Darko, Kiciak, Badu, Grigorov, Fleck and Osei 16 The goal of these studies were to investigate the capability of the QA procedure including the dosimetry equipment and the comparison technique (gamma analysis) to detect potential errors that could occur due to mis-calibration or mechanical errors of the treatment machine. Kim et al. investigated the change in gamma passing rate (%GP) after introducing three types of MLC misalignments (closing, opening and shifting) by 0.25, 0.5, 1 and 2 mm.Reference Kim, Park, Kim, Kim, Ye and Park 1 They found that a global gamma criterion of 2%/1 mm and a low-dose threshold of 10% with passing rates of 90 and 80% was able to detect open and close MLC misalignments of 0.5 mm and shift misalignments of 1 mm using MapCHECK2TM detector array (Sun Nuclear Corp. ) and EBT2TM film (Ashland Inc., Covington, KY, USA), respectively, for patient-specific VMAT QA for lung and localised spine metastasis stereotactic body radiation therapy (SBRT).Reference Kim, Park, Kim, Kim, Ye and Park 1 Heilemann et al. introduced similar MLC positioning errors and showed that using a global gamma index criterion of 2%/2 mm and a 10% low dose threshold with a %GP of 90%, MLC open misalignment of 1 mm, MLC closed misalignment of 0.5 mm for prostate and 1.0 mm for head and neck cases, and MLC shift misalignment of 3 mm can be detected.Reference Heilemann, Poppe and Laub 4 Their measurements were performed using the OCTAVIUSTM 2D-Array (PTW, Freiburg, Germany) and the Delta4 TM devices.Reference Heilemann, Poppe and Laub 4 Fredh et al. used Delta4 TM, seven29TM 2D-array (PTW) and the OCTAVIUSTM system, MatriXXEvolution TM ion chamber array (IBA Dosimetry GmbH, Schwarzenbruck, Germany) and COMPASSTM (IBA Dosimetry GmbH) system, and aS1000TM EPID (Varian Medical Systems) and EpiqaTM software (EPIdos, Bratislava, Slovakia) system.Reference Fredh, Scherman, Fog and Munck af Rosenschold 5 They introduced errors that included 3% increase in number of monitor units, widening of the MLC bank by 2 and 4 mm and rotation of the collimator by 2° and 5°. They found that by using 2%/2 mm criteria and %GP of 95% the Delta4 TM detected 15 of 20 errors, OCTAVIUSTM detected 8 of 20 errors, COMPASSTM detected 8 of 20 errors and EpiqaTM detected 20 of 20 errors.Reference Fredh, Scherman, Fog and Munck af Rosenschold 5
Some studies have questioned the correlation between the gamma evaluation pass rates and the deviations observed in the dose volume histograms (DVH).Reference Kim, Park, Kim, Kim, Ye and Park 1 , Reference Liang, Liu, Zhou, Yin and Wu 3 – 9 Coleman et al. studied the sensitivity of gamma passing rates by setting a threshold of 3 and 5% mean absolute dose error on planning target volume (PTV) D 95 and PTV D mean, respectively, and a 5% threshold on dose error in bladder and rectum D mean DVH metrics.Reference Coleman and Skourou 6 They concluded that there is not a consistently strong correlation between gamma index and DVH metrics obtained using the ArcCHECKTM and 3DVHTM (Sun Nuclear Corp.) software for QA.Reference Coleman and Skourou 6 Heilemann et al. Reference Heilemann, Poppe and Laub 4 found a weak correlation between gamma-index passing rates and dosimetric effects on the DVH. They considered an increase of mean dose to PTV by more than 2% as clinically significant which was also suggested by Oliver et al. Reference Oliver, Gagne, Bush, Zavgorodni, Ansbacher and Beckham 7 . Kim et al. observed no correlations between dose–volumetric changes and gamma passing rates in lung and spine SBRT by accepting a 0.25 mm MLC closing and opening misalignment and 1 mm of shift misalignment as clinically tolerable.Reference Kim, Park, Kim, Kim, Ye and Park 1 Stasi et al. found a weak correlation between the %GP and the absolute percentage dose differences that were obtained by MapCheckTM and 3DVHTM software for IMRT QA while considering DVH errors higher than 5 or 3% as clinically significant change.Reference Stasi, Bresciani, Miranti, Maggio, Sapino and Gabriele 8
In this study we investigated the sensitivity of the gamma index analysis to detecting MLC positioning errors using the Varian TrueBeamTM EPID and ArcCHECKTM for patient-specific prostate VMAT quality assurance. Simulated positioning errors with a mean value of 0.25, 0.5, 1 and 2 mm and a standard deviation of 0.1 mm were applied on all active MLCs (randomly as either an opening or closing) and on randomly selected 25, 50, 75 and 100% of control points. The clinical impact of the MLC positioning errors introduced on the dose distributions of the PTV and organs at risk (OAR) were evaluated and compared to the change in %GP to determine possible correlation between MLC positioning errors with PTV dose coverage and doses to the OAR.
MATERIALS AND METHODS
Linear accelerator and treatment planning software
VMAT plans of seven previously treated prostate cancer patients were randomly selected for this study. Treatment planning was performed using the EclipseTM TPS (Varian Medical Systems) version 11 and by means of the analytical anisotropic algorithm with a grid size of 0.25 cm. The plans were generated using 6 MV photon beams with two arc fields and a prescribed dose of 7,800 cGy in 39 fractions for the treatment of intact prostate. All plans were delivered using a Varian TrueBeamTM linear accelerator equipped with a Millennium 120 MLC. Daily QA is performed on the linear accelerator to ensure consistency in output, symmetry and flatness.
Measuring detector arrays
Two detector arrays, the Varian EPID and Sun Nuclear ArcCHECKTM were used for all QA measurements. The EPID is an amorphous silicon (aS1000) panel with an active area of 40×30 cm2 and a 0.392 mm pixel resolution. The EPID was calibrated according to the vendor’s specifications. The EPID response was scaled such that 1 calibrated unit (CU) corresponds to 100 MU delivered by a 10×10 cm2 open field at 100 cm source-imager-distance (SID). Measurements were all performed at the same SID of 100 cm as used during the absolute calibration of the imager with no additional build up on the imager. Varian Portal Dosimetry software version 11 was used for the analysis.
The ArcCHECKTM system contains 1,386 n-Si diode array arranged in a helical pattern along a cylindrical phantom, resulting in semi-3D dose measurements. The detectors are 1 cm apart and form 21 helical continuous rings with 66 detectors on each ring. Measurements were performed with the cavity plug inserted. The calibrations were performed according to the manufacturer’s instructions and absolute dose calibration was performed prior to measurements. A QA plan with four 10×10 cm2 fields at 90°, 0°, 270° and 180° gantry angles was acquired and compared with the verification plan prior to measurements to make sure that the devise is properly setup and consistent during each measurement. SunNuclear SNC Patient software version 6.4.1 was used for the acquisition and analysis of the measurements.
MLC leaf position errors
An in-house PythonTM program was developed to deliberately introduce specific errors in the MLC positions of each VMAT plan. For each VMAT plan, the original plan dicom file was exported from the TPS to the PythonTM program. The program is used to modify the MLC positions by introducing random positioning errors with a mean of 0.25, 0.5, 1 and 2 mm and a standard deviation of 0.1 mm (by sampling a Gaussian function) on all active MLCs. The errors were randomly applied on 25, 50, 75 and 100% of the 177 control points. For purposes of clarity in this paper, an error with a mean of 0.25 mm that was applied on 25% of the control points will be referred to as ‘0.25 mm-25%.’ The direction (open or close) of the change in the MLC positions was also randomly selected. In situations where leaf collision will occur as a result of the simulated misalignment, the leaf pair distance (leaf gap) is set to 0.5 mm which is the smallest possible leaf gap; 16 plans based on the simulated MLC position errors were generated from each original plan. In addition to the seven original patient plans, a total of 119 plans were measured on both detector systems.
Gamma evaluation
The improved gamma calculation algorithm within the ARIA Portal Dosimetry Review workspace in the Eclipse TPS was used for the EPID analysis. The global gamma calculation algorithm has been replaced by the improved gamma calculation algorithm in the Varian Portal Dosimetry version 11.0. In the previous algorithm (global), when searching for distance to agreement, the system would only consider integer pixel positions around the pixel being evaluated. According to Varian Portal Dosimetry reference guide 9 ‘this sampling limitation may result in an overestimation of the gamma value at the evaluation point.’ Thus the improved option allows the evaluation to interpolate between neighbouring pixels when searching.
We used the improved and global gamma calculation algorithms with a low dose threshold of 10% in the EPID and ArcCHECK, respectively. The gamma index was calculated using the absolute dose for both devices. Several gamma criteria of 3%/3 mm, 3%/2 mm, 3%/1 mm, 2%/2 mm, 2%/1 mm and 1%/1 mm were considered in order to investigate the change in gamma passing rates for each criterion as a function of the magnitude of the MLC error. In this study, a multi gamma criteria technique was also proposed and the sensitivity of both the EPID and ArcCHECK detectors in detecting specific MLC poisoning error using this technique was investigated. In this technique, the change in the %GP was considered for all the gamma criteria (3%/3 mm, 3%/2 mm, 3%/1 mm, 2%/2 mm, 2%/1 mm and 1%/1 mm) as a group rather than as individual criterion.
Impact of MLC error on dose distribution
The modified treatment plans were imported back into the TPS and the dose distribution calculation. Table 1 represents the volume and the dose constrains sets used for the evaluation of the dose distribution to the target and critical organs. The mean dose values of the PTV were also determined for each plan.
Treatment planning objectives
Data analysis
Three types of analysis were performed to study the sensitivity of gamma index to detect the errors that were introduced in the MLC positions. For the first set of analysis, the measured plans (with and without error) were compared with the predicted original plan (referred to as ‘PR-MS’) which is the regular procedure for QA. The second analysis was done comparing the predicted plans with the induced errors and their corresponding error free original predicted plans (referred to as ‘PR-PR’). This was done to theoretically (with no inherent and measurement uncertainties) compare the capability of the EPID and the ArcCHECK to detect errors using the change in %GP. The last set of analysis was performed comparing the measured plans with error to that without error (referred to as ‘MS-MS’). The purpose of doing this was to compare the sensitivity of %GP to the introduced errors using the two devices with the presence of inherent uncertainties of the delivery and detection. For the gamma analysis and for all three sets of analysis that is PR-PR, MS-MS and PR-MS, the reference dose distribution was always the original (error free) plan or measurement. For the PR-MS analysis, the reference dose was the predicted original plan that was compared against the measured (with and without error) plans.
RESULTS AND DISCUSSION
The means and standard deviations (SD) of %GP for each gamma criterion were plotted as a function of the magnitude of simulated error in MLCs positioning and the percentage of control points affected by the error for the measured (PR-MS) data (Figure 1) and the theoretical (PR-PR) data (Figure 2). The average and SD of %GP for multi gamma criteria technique for PR-MS and PR-PR data are shown in Figures 3 and 4, respectively. The results of the gamma passing rates for MS-MS data were very similar to those of PR-PR data and therefore, they are not presented. Table 2 shows the mean and standard deviation in percentage change of PTV mean dose, V 95%, V 100%, conformity index (V 95%/V tot), rectum V 50 Gy, V 65 Gy, V 75 Gy, bladder V 65 Gy, V 75 Gy, V 80 Gy, and the maximum dose to the femoral heads for the modified plans from their original plans. The per cent difference in volume or dose between the plan with simulated errors and the plan without errors is calculated as (D Error−D Original)/D Original. Therefore, a negative value means a reduction in dose (or volume) from its original value.
The PR-MS gamma analysis of plans with 0 mm (original), 0.25, 0.5, 1 and 2 mm of simulated error in multileaf collimator (MLC) leaves on (a) and (b) 25%, (c) and (d) 50%, (e) and (f) 75%, and (g) and (h) 100% of control points using the electronic portal imaging device (EPID) (left column) and ArcCHECK (right column). The error bars represent the standard deviation of 7 data points. Abbreviation: GP, gamma passing.
The PR-PR gamma analysis of plans with 0.25, 0.5, 1 and 2 mm of simulated error in multileaf collimator (MLC) leaves on (a) and (b) 25%, (c) and (d) 50%, (e) and (f) 75%, and (g) and (h) 100% of control points using the electronic portal imaging device (EPID) (left column) and ArcCHECK (right column). The error bars represent the standard deviation of 7 data points. Abbreviation: GP, gamma passing.
The PR-MS multi gamma criteria analysis of plans with 0 mm (original), 0.25, 0.5, 1 and 2 mm of simulated error in multileaf collimator (MLC) leaves on (a) and (b) 25%, (c) and (d) 50%, (e) and (f) 75%, and (g) and (h) 100% of control points using the electronic portal imaging device (EPID) (left column) and ArcCHECK (right column). The error bars represent the standard deviation of 7 data points. Abbreviation: GP, gamma passing.
The PR-PR multi gamma criteria analysis of plans with 0.25, 0.5, 1 and 2 mm of simulated error in multileaf collimator (MLC) leaves on (a) and (b) 25%, (c) and (d) 50%, (e) and (f) 75%, and (g) and (h) 100% of control points using the electronic portal imaging device (EPID) (left column) and ArcCHECK (right column). The error bars represent the standard deviation of 7 data points. Abbreviation: GP, gamma passing.
The mean and SD of percentage change in dose–volumetric indicators
Abbreviation: PTV, planning target volume; Conf. Ind., conformity index.
Two techniques were used in order to examine the sensitivity of EPID and ArcCHECK to detect the errors that were introduced in the MLC positions for PR-MS and PR-PR data using the single and multi-gamma criteria techniques. For the first technique (single-gamma criterion), two tailed t-tests were performed between the %GPs of the original plans of seven patients with those of the modified plans for each gamma criterion. This was done as a function of the control points that were affected by the simulated error (25, 50, 75 and 100%). The smallest error in magnitude (0.25, 0.5, 1 or 2 mm) that caused the %GPs of the modified plans to be significantly different than those of the original plans with a confidence level of 95% (p≤0.05), were recorded and presented in Table 3 (PR-MS data) and Table 4 (PR-PR data). For the second technique using the multi-gamma criteria, analysis of variance two-factor with replication test was performed between the gamma passing rates of seven criteria for seven original and modified plans. The smallest error that caused the seven %GPs of the modified plans to be significantly different than those of the original plans was considered as the smallest detectible error for each detector. The results are presented in Table 5. The smallest detectable error for the PR-MS and PR-PR data using the multi gamma criteria technique is shown in Table 5.
The smallest detectable error for the PR-MS data
Abbreviation: EPID, electronic portal imaging device.
The smallest detectable error for the PR-PR data
Abbreviation: EPID, electronic portal imaging device.
The smallest detectable error for the PR-MS and PR-PR data using the multi-gamma criteria technique
Abbreviation: EPID, electronic portal imaging device.
The measured (PR-MS) results of Figures 1 and 3 and Tables 3 and 5 demonstrate that the Varian TrueBeamTM EPID has a relatively higher sensitivity in detecting errors that were introduced to MLC positions. This could be partly due to the higher detector resolution for the EPID compared with the ArcCHECKTM and also due to the difference in detector layout in the two systems.Reference Liang, Liu, Zhou, Yin and Wu 3 Liang et al. also found the ArcCHECK to be less sensitive to MLC position error compared to the EPID.Reference Liang, Liu, Zhou, Yin and Wu 3 Using the multi-gamma criteria technique (Table 5), the EPID can detect errors as small as 0.5 mm when 100% or even 75% of control points are affected. The smallest detectable error using the ArcCHECK is 2 and 1 mm on 75 and 100% of control points, respectively. The error 0.5 mm-75%, which is the smallest detectable error using the EPID, corresponds to a mean percentage change of −0.33±0.86, 0.39±0.65 and −0.11±0.54 for the minimum, maximum and mean PTV dose, respectively. The error 2 mm-75%, which is the smallest detectable error using the ArcCHECK, corresponds to a mean percentage change of −4.26±2.95, 2.56±2.15 and −0.65±2.15 for the minimum, maximum and mean PTV dose, respectively. The theoretical (PR-PR) results of Figures 2 and 4 and Tables 4 and 5 show no noticeable difference between the sensitivity of the two detectors to MLC misalignments. This could be because the predicted plans calculated by the planning system for both detectors have the same resolution of 0.78×0.78 mm2.
The measured results of Tables 3 and 5 show that smaller errors could be detected using the multi-gamma criteria technique in comparison with the single-gamma criterion technique when using the EPID. For example, when 25, 50, 75 and 100% of the control points were affected by the MLC position error, the smallest detectable errors were 2, 1, 0.5 and 0.5 mm using the multi-gamma criteria technique and 2, 1, 1 and 1 mm using the 2%/2 mm criterion, respectively. However, for the ArcCHECK, the smallest detectable errors obtained using the multi-gamma criteria technique were in general similar to those obtained using the single-gamma criterion technique for the PR-MS data.
Comparing the theoretical results of single- (Table 4) and multi- (Table 5) gamma criteria techniques shows a major improvement in detecting MLC position misalignment using a multi-gamma criteria method for both QA systems. The likely reason for the improvement in detecting MLC position misalignment with the multi-gamma criteria could be because the method simultaneously takes into account the low sensitivity but a small variability and the high sensitivity but high variability of the different gamma criteria. Each individual criterion has its own limitations. A criterion such as 3%/3 mm has a low sensitivity but a small variability, whereas a criterion such as 1%/1 mm has a high sensitivity but suffers from high variability. A multi-gamma criteria could potentially be better in patient-specific QA than using a single-gamma criterion. As discussed by previous studies, the 3%/3 mm criterion is the least sensitive criterion to the introduced errors.Reference Kim, Park, Kim, Kim, Ye and Park 1 , Reference Liang, Liu, Zhou, Yin and Wu 3 – Reference Nelms, Chan and Jarry 13 None of the errors were detected with EPID when 25% of control points had the MLC errors. ArcCHECK was not capable of detecting any of the introduced errors when half of the control points were affected by the error when using the 3%/3 mm criterion.
Table 2 shows that the rectum V 75 Gy, bladder V 80 Gy and maximum dose to the femoral heads changes by −1.3±3.7%, 0.4±2.4% and 1.0±2.9%, respectively, due to the largest error, 2 mm-100%. However, all the modified plans met the criteria that were set for OAR. In this study, we considered a 3% deviation in PTV V 95% (with no tolerance on constraint sets for OAR and the conformity index) as clinically significant change in the plan. The errors which did not meet these criteria and caused a clinically significant change in the original plans were 2 mm-75% and 2 mm-100%. Figures 1 and 3 and Tables 3 and 5 show that both of these errors are detectable using both detectors.
CONCLUSIONS
The present study show that the Varian TrueBeam EPID dosimetry demonstrate a higher sensitive in detecting MLC positioning errors compared with the ArcCHECK for pre-treatment patient-specific QA. In addition, using a multi-gamma criteria approach seems to improve the sensitivity of the EPID. Further studies will be required to investigate the sensitivity of the two detectors in detecting MLC positioning errors in larger fields and with higher degrees of MLC modulation.
ACKNOWLEDGEMENT
The authors would like to acknowledge with much gratitude the financial support through research grant provided by the TELLUS Ride for Dad and the Prostate Cancer Fight Foundation for this study.
