Transit and non‐transit 3D EPID dosimetry versus detector arrays for patient specific QA

Abstract Purpose Despite their availability and simplicity of use, Electronic Portal Imaging Devices (EPIDs) have not yet replaced detector arrays for patient specific QA in 3D. The purpose of this study is to perform a large scale dosimetric evaluation of transit and non‐transit EPID dosimetry against absolute dose measurements in 3D. Methods After evaluating basic dosimetric characteristics of the EPID and two detector arrays (Octavius 1500 and Octavius 1000SRS), 3D dose distributions for 68 VMAT arcs, and 10 IMRT plans were reconstructed within the same phantom geometry using transit EPID dosimetry, non‐transit EPID dosimetry, and the Octavius 4D system. The reconstructed 3D dose distributions were directly compared by γ‐analysis (2L2 = 2% local/2 mm and 3G2 = 3% global/2 mm, 50% isodose) and by the percentage difference in median dose to the high dose volume (%∆HDVD 50). Results Regarding dose rate dependency, dose linearity, and field size dependence, the agreement between EPID dosimetry and the two detector arrays was found to be within 1.0%. In the 2L2 γ‐comparison with Octavius 4D dose distributions, the average γ‐pass rate value was 92.2 ± 5.2%(1SD) and 94.1 ± 4.3%(1SD) for transit and non‐transit EPID dosimetry, respectively. 3G2 γ‐pass rate values were higher than 95% in 150/156 cases. %∆HDVD 50 values were within 2% in 134/156 cases and within 3% in 155/156 cases. With regard to the clinical classification of alerts, 97.5% of the treatments were equally classified by EPID dosimetry and Octavius 4D. Conclusion Transit and non‐transit EPID dosimetry are equivalent in dosimetric terms to conventional detector arrays for patient specific QA. Non‐transit 3D EPID dosimetry can be readily used for pre‐treatment patient specific QA of IMRT and VMAT, eliminating the need of phantom positioning.


| INTRODUCTION
The major recent advances in modern external beam radiation therapy, particularly since the arrival of VMAT, demand dosimetric methods and tools to perform patient specific QA in multiple planes or in 3D. 1,2 The only types of dosimeters currently capable of measuring full-3D dose distributions are polymer gel dosimeters and radiochromic detectors. [3][4][5][6] A more popular option is the use of pseudo 3D dosimeters consisting of a multidimensional detector array capable of reconstructing a portion or the whole 3D dose distribution within a phantom. Commercially available detector arrays, such as Arc-CHECK (Sun Nuclear Corp., Melbourne, FL), Octavius 4D (PTW, Freiburg, Germany), and Delta4 (Scandidos AB, Upsala, Sweden), compare measured to planned 3D dose distributions. Several studies have already investigated the dosimetric characteristics and performance of these devices. [7][8][9][10][11][12][13][14][15] Although initially developed for patient setup verification, Electronic Portal Imaging Devices (EPIDs) also have useful dosimetric characteristics as summarized in review articles. 16,17 Some EPID-based approaches already allow for patient specific QA by reconstructing 3D dose distributions within the patient or phantom anatomy. In 3D non-transit EPID dosimetry, the dose is determined within the patient or phantom based on in air or fluence EPID measurements, that is, without an attenuating medium between the source and the detector. [18][19][20] In 3D transit EPID dosimetry, the dose is reconstructed within the patient or phantom based on EPID measurements acquired behind the patient or phantom. [21][22][23][24][25] A clear advantage of EPIDs is their availability which makes them suitable for large scale implementations. When used for patient specific QA, non-transit 3D EPID dosimetry is easier and faster than transit 3D EPID dosimetry because it eliminates the need to position a phantom. Furthermore, in cases where non-transit EPID dosimetry allows for 3D dose reconstruction within the patient anatomy, the need for phantom re-planning is also eliminated.
One of the main reasons why EPIDs have not yet replaced detector arrays for patient specific 3D dose verification is that commercial solutions are not widely available yet. Furthermore, there is a lack of studies validating EPID dose distributions against absolute dose measurements in 3D. This is typically due to a combination of the cumbersome work required and software challenges. As a result, these studies usually comprise only a limited number of cases. 19,20,26 The comparison with ion chamber measurements is feasible only in a limited number of points and hence insufficient for the validation of complex IMRT and VMAT plans. Consequently, 3D EPID-based dosimetry solutions generally end up being validated against the TPS.
The aim of this study is to assess the accuracy of transit and non-transit 3D EPID dosimetry versus detector arrays. After evaluating basic dosimetric characteristics of the EPID and two detector arrays (Octavius 1500 and Octavius 1000 SRS ), 3D dose distributions for 68 VMAT arcs, and 10 IMRT plans were reconstructed using EPID dosimetry and the Octavius 4D system. The reconstructed 3D dose distributions were directly compared by γ-analysis and by differences in the median dose to the high dose volume. The clinical impact of replacing the Octavius 4D system by EPID dosimetry for patient specific pre-treatment QA was also evaluated.

2.A | Equipment
Measurements were performed on a VersaHD linear accelerator (Elekta AB, Stockholm, Sweden) equipped with an Agility MLC (5 mm leaf width). Also available was the MV imaging acquisition software iViewGT™ (Elekta AB, Stockholm, Sweden) with an amorphous silicon EPID (PerkinElmer XRD 1642 AP) which is situated at 160 cm distance from the linac target. The EPID panel has a detection area of 41 × 41 cm 2 (1024 × 1024 pixels). Treatment plans with all available energies (6 MV, 10 MV, 6 MV FFF, and 10 MV FFF) were generated with Pinnacle V9.10 (Philips Medical Systems, Eindhoven, The Netherlands).

2.B | Detector arrays
The detectors used in this study were the Octavius 1500 and Octavius 1000 SRS 2D arrays (PTW, Freiburg, Germany). The 1500 array has an active area of 27 × 27 cm 2 with a center-to-center detector spacing of 7.1 mm. The 1000 SRS array is typically used for small target plans due to its smaller active area and higher resolution. In this array, the center-to-center detector spacing is 2.5 mm in an area of 5.5 cm × 5.5 cm around the center and 5 mm in the remaining outer area (up to 11 cm × 11 cm). In both cases the reconstructed dose cube has a 2 mm grid. A detailed characterization of both devices can be found elsewhere. 27,28 The arrays are inserted in the Octavius 4D phantom which is a cylindrical phantom made of water-equivalent white polystyrene (RW3) with a slot to hold the detector array.
An inclinometer is placed on the gantry, this is connected to a control unit that sends the movement information to the phantom. This allows the detector to rotate synchronously with the gantry ensuring that the detector remains perpendicular to the beam at all times during delivery. Based on these measurements, 3D dose distributions can be reconstructed within the Octavius 4D phantom geometry using the Verisoft 7.1 software (PTW, Freiburg, Germany). For both 2D detector arrays, a dose calibration prior to measurements is performed by delivering a 10 × 10 cm 2 field per beam energy and providing the system with the expected dose.

2.C | 3D transit EPID dosimetry
The commercially available iViewDose system (Elekta AB, Stockholm, Sweden) is a transit EPID dosimetry solution that uses EPID images acquired behind a patient in combination with CT data to reconstruct 3D dose distributions within the patient anatomy. Although intended for in vivo dose verification, iViewDose can also be used for pretreatment verification in combination with a phantom. Full details regarding the dose reconstruction can be found in the original design of the algorithm. 21,29 For IMRT verification, the iViewGT acquisition software (Elekta AB, Stockholm, Sweden) averages the total signal of all EPID frames between beam-on and beam-off into one accumulated portal image. The reconstructed 3D dose distributions of all beams are then summed to obtain the 3D dose distribution of the IMRT fraction. For VMAT verification, cine-mode image acquisition is used and separate EPID frames are continuously being acquired during delivery. Each recorded frame is associated with a gantry angle. The reconstructed 3D dose distributions of all frames are then summed to obtain the 3D dose distribution of the total VMAT arc.
In order to determine the accuracy of transit EPID dosimetry, use was made of a unique experimental setup. Transit EPID images were acquired behind an Octavius 4D phantom with an homogeneous insert, that is an entirely RW3-filled phantom. A CT scan of The whole process takes about 3 h per energy. iViewDose uses a simple 0D couch attenuation model that also has to be comissioned for each energy. 30

2.D | 3D non-transit EPID dosimetry
A research version of the same iViewDose package allows for nontransit EPID dosimetry in 3D. The non-transit algorithm uses in air EPID measurements in combination with CT data to determine a "virtual" transit primary portal dose distribution at the EPID level.
Virtual transit primary portal dose distributions predict the transit primary portal dose distributions that would be measured behind a patient without changes in anatomy. In IMRT reconstructions, the non-transit algorithm uses accumulated in air portal images to estimate the accumulated "virtual" transit primary portal dose distribution corresponding to each field. In VMAT reconstructions, the nontransit algorithm uses in air portal arcs to estimate the "virtual" transit primary portal dose distribution of each VMAT frame. These virtual transit primary portal dose distributions are used in combination with patient CT data to reconstruct 3D dose distributions directly within the patient anatomy. This method is therefore very useful for patient specific pretreatment QA since it provides information about the dose to be delivered to the patient, including potential machine and planning errors. Note that apart from the determination of the primary portal dose distribution, the transit and non-transit dose reconstruction algorithms are identical. In order to determine the accuracy of non-transit EPID dosimetry in this study, in air EPID measurements were used in combination with the CT data set of the adapted Octavius 4D phantom to reconstruct non-transit 3D dose distributions within this phantom geometry. Full details regarding the non-transit dose reconstruction engine can be found elsewhere. 19

2.E | Dosimetric characterization
The dose rate dependency was evaluated, for each combination of measurement configuration and energy shown in Fig. 1, by irradiating a total of 100 MU for a 10 × 10 cm 2 field at five different dose rates (100%, 50%, 25%, 12.5%, and 6% of the maximum dose rate).
For each energy, the ratio of the reading to MU delivered was calculated and normalizsed to the maximum dose rate. The dose linearity was studied over the range 5-1000 MU for flattened beams and 10-1000 MU for unflattened beams. The ratio of the reading to MU delivered was calculated and normalised to 100 MU in each case. All measurements were made at gantry 0°. Note that the Octavius 1000 SRS presents a non-negligible dependence on dose rate as well as on field size. 31,32 The manufacturer of the Octavius 1000 SRS array recommends to cross-calibrate the detector at a 4 × 4 cm 2 field size to mitigate the overall effect of the field size dependence. In our clinic, the detector is calibrated at a 10 × 10 cm 2 field size and an in-house implemented correction is applied to each measurement frame prior to the reconstruction. In this correction, the actual effective field size and dose rate are first determined from the data of each measurement frame, and then a correction factor for that field size and also for the dose rate is applied on a per-frame basis. The correction factors are empirically determined by comparing measurements performed with the detector array to reference data.

3.C | IMRT and VMAT treatments
The results of the direct comparison between the Octavius 4D system and EPID dosimetry are presented in Table 1 The highest 2L2 γ-mean value (0.72) was found for a VMAT head-and-neck case, indicated by the blue circled data points in Fig. 4. This case presented also the lowest 3G2 γ-pass rate (93.5%).
Graphical data corresponding to this analysis are exhibited in Fig. 5a.
As can be seen in the profiles, EPID dosimetry showed a better agreement with the TPS than Octavius 4D. The red circled data points in Fig. 4 correspond to a 6MV FFF VMAT stereotactic brain plan which presented one of the largest HDV D50 differences (−2.8%). The underdosage with EPID dosimetry is further displayed in the zoomed parts of Fig. 5b. However, the 2L2 γ-mean value for this comparison was 0.56 which illustrates the insufficiency of γ-evaluation alone to compare dose distributions in high dose gradient volumes. In this case, Octavius 4D presented a better agreement with the TPS than EPID dosimetry. For completeness, an IMRT lung case is displayed in Fig. 5c where EPID dosimetry and Octavius showed a better agreement among themselves than with the TPS. angles are sampled we expect that the spatial resolution of the 1500 detector will cause problems in this particular case. The other VMAT head-and-neck measurements included in this study corresponded not only to less modulated cases but were also delivered as full arcs

3.D | Comparison with TPS: alert classification rate
where the effects of the lack of spatial resolution are expected to smooth out. There were three more cases, circled in gray in Fig. 6a, where a poor agreement with the plan was obtained with both Octavius 4D and EPID dosimetry. These cases are further discussed in the discussion section.

transit EPID non-transit EPID
F I G . 4. Scatter plots displaying results of the direct comparison between 3D dose distributions reconstructed with Octavius 4D and the corresponding 3D dose distributions reconstructed with transit and non-transit EPID dosimetry using (a) 2L2 γ-mean values, (b) 3G2 γ-pass rate values, and (c) percentage difference in median dose to the high dose volume %ΔHDV D50 (Octavius 4D used as reference). The VMAT cases in (c) are arranged in increasing order of analyzed volume. The circled data points correspond to a 6MV head-and-neck VMAT plan that showed the largest 2L2 γ-mean value (blue), to a 6MV FFF VMAT stereotactic brain plan that presented one of the largest deviations in HDV D50 (red) and to a 10MV IMRT lung case (black).
corrections for these effects. The results of this study indicate an excellent dosimetric agreement between the 3D dose distributions reconstructed with EPID dosimetry and Octavius 4D. Average 2L2 γmean values below 0.6 suggest that, on average, a dose difference of~1.2% is expected in low-dose gradient regions or a DTA of 2 mm in steep dose gradient regions. The average 3G2 γ-pass rate value found for the 10 IMRT plans and 68 VMAT arcs was 98.8%.
For comparison, a recent study comparing 3D dose distributions reconstructed with Octavius 4D and with non-transit EPID dosimetry reported 3G3 γ-pass rates of 95% and 93% for one IMRT prostate plan and one VMAT head-and-neck plan, respectively. 20 Also, in the initial investigation of the non-transit dose reconstruction algorithm used in this study, 19 an average 3G3 γ-pass rate value of 98.5% was reported for 5 IMRT and 5 VMAT plans. In another recent study, the EPID-reconstructed 3D dose distribution for one VMAT lumbar spine vertebra was compared with gel dosimetry reporting a 3G3 γpass rate value of 93%. 26 As can be seen from the 2L2 γ-results in Table 1   possibly due to the same spatial resolution issues as in the case circled in blue in Fig. 6. In the plan calculated with 2°gantry spacing, most of the dose is delivered in two peaks within a 20°degree gantry angle value range. Profiles of the TPS versus any of the detectors presented low peaks across the high dose region and high valleys across the low dose region. Unfortunately, an increase in the gantry spacing resolution to 1°is not allowed by the planning system.
The transit 3D EPID dosimetry system is clinically used for in vivo dose verification of all our external photon beam treatments.
The high accuracy shown by transit 3D EPID dosimetry in this study confirms the reliability of the method. In vivo dose verification presents, however, extra challenges and hence, additional uncertainties.
First, the parameters of the dose reconstruction algorithm are determined using water-based scatter correction kernels and consequently, the model does not account for tissue inhomogeneities accurately. Therefore, for dose verifications of sites involving (large) tissue heterogeneities, the in aqua vivo approach must be used. 38 Second, the dose delivered to the patient during treatment is affected by patient setup variations and patient anatomical changes. 39 Last, another limitation of the algorithm is that it uses the planning CT as patient anatomy model for dose reconstruction. To better cope with the challenge of patient variations, more accurate results are expected once the daily patient anatomy is used in the reconstruction.

| CONCLUSION
Transit and non-transit EPID dosimetry are equivalent in dosimetric terms to conventional detector arrays for patient specific QA. Nontransit 3D EPID dosimetry can be readily used for pretreatment patient specific QA of IMRT and VMAT, eliminating the need of phantom positioning.

CONF LICT OF I NTEREST
Our department licenses software for portal dosimetry to Elekta Oncology Systems Ltd. The authors are responsible for the content and writing of the paper.