Abstract
We propose a method to incorporate independent verification of gantry angle for electronic portal imaging device (EPID)-based pre-treatment quality assurance (QA) of clinical volumetric modulated arc therapy (VMAT) plans. Gantry angle is measured using projections in the EPID of a custom phantom placed on the couch and the treatment plan is modified so as to be incident on the phantom with a portion of the beam that is collimated in the clinical plan. For our implementation, collimator and couch angles were set to zero and the inferior jaw and two most inferior multi-leaf collimator pairs were opened for the entire QA delivery. A phantom containing five gold coils was used to measure the gantry rotation through which each portal image was acquired. We performed the EPID QA for ten clinical plans and evaluated accuracy of gantry angle measurement, scatter incident on the imager due to the phantom, inter-image pixel linearity and inter- and intra-image noise. The gantry angle could be measured to within 0.0 ± 0.3° for static gantry and 0.2 ± 0.2° for arc acquisitions. Scatter due to the presence of the phantom was negligible. The procedure was shown to be feasible and adds gantry angle to the treatment planning parameters that can be verified by EPID-based pre-treatment VMAT QA.
Export citation and abstract BibTeX RIS
General scientific summary Radiotherapy treatment plans have increased in complexity so that beam multi-leaf collimators (MLCs), dose rate, and gantry angle can all be modified simultaneously during delivery. Such complex plans are referred to as volumetric modulated arc therapy (VMAT), and must be verified prior to treatment. Electronic portal imaging devices (EPIDs) mounted on clinical linear accelerators have become useful imaging and quality assurance (QA) tools for radiation therapy. We proposed and evaluated a method to incorporate independent verification of gantry angle for EPID based pre-treatment QA of modified VMAT plans using projections through a custom phantom. The method adds the gantry angle to the treatment planning parameters that can be verified by EPID based pre-treatment VMAT QA and has potential for application to commissioning and routine machine QA.
1. Introduction
Plan-specific intensity modulated radiation therapy (IMRT) quality assurance (QA) is performed in many radiotherapy centers to ensure accurate treatment delivery (Ezzell et al 2003). Modern linear accelerators are often equipped with an electronic portal imaging device (EPID) for image guidance, which has also been utilized for IMRT QA. IMRT QA using an EPID has the advantage of efficient delivery and analysis without requiring a separate measurement device. Meanwhile, volumetric modulated arc therapy (VMAT) is becoming increasingly common in clinical settings and serves to decrease treatment duration. Some recent studies have investigated extending current EPID IMRT QA to VMAT by acquiring an integrated image during static or arc QA delivery of the clinical VMAT plan (Iori et al 2007, 2010), or dividing each VMAT arc into multiple sub-arcs for which an integrated image is acquired per sub-arc (Nicolini et al 2008). Another technique verifies multi-leaf collimator (MLC) positions detected from cine images acquired at a high temporal resolution during the VMAT arc delivery (Bakhtiari et al 2011). EPID-based QA verifies the treatment delivery parameters as opposed to the dose distribution. As in traditional EPID-based IMRT QA, these VMAT QA techniques verify MLC positions as a function of delivered MU either directly or by a pixel-intensity-based comparison.
VMAT has a level of complexity beyond traditional IMRT due to additional motion of the gantry, which is also synchronized with the MLC motion. Since variable gantry rotation speed during VMAT delivery is a new change in the treatment delivery, it is necessary to verify that the gantry rotates according to the treatment plan during delivery and its trajectory must be verifiable. As the MLC positions, dose rate and gantry angle can all modulate simultaneously, EPID-based pre-treatment QA for VMAT would be more comprehensive if gantry angle was also verified along with MLC positions and delivered MU. By design, the EPID cannot detect the gantry rotation directly due to its fixed position with the gantry. The most straightforward solution when MV cine imaging is used is to compare the expected gantry angle with the gantry angle supplied by the vendor and recorded in the digital imaging and communications in medicine (DICOM) header of the cine image. However, studies have shown the DICOM recorded gantry angle for some vendor systems to have poor precision (Ansbacher et al 2010). Another possibility would be the use of a mechanical inclinometer. While a mechanical inclinometer would have the advantage of providing an independent measurement of gantry angle, it may be subject to effects from time delay and inertia and would require synchronization with a linear accelerator or with the acquired EPID images. In this study, we propose and investigate the feasibility of using projections in the EPID image of a custom-designed phantom to detect the gantry rotation through which each image is acquired during pre-treatment VMAT QA. By encoding the gantry angle directly into the acquired image, no synchronization of the gantry angle measurement with the accelerator or imaging system is required. Furthermore, this technique is potentially applicable to many of the proposed EPID-based VMAT QA techniques discussed above and adds gantry angle to the treatment delivery parameters that can be verified by EPID-based pre-treatment QA for VMAT.
2. Methods and materials
2.1. Concept
The proposed QA procedure consists of the following steps: (1) modification of the clinical VMAT plan for QA delivery, (2) delivery of the QA plan with continuous acquisition of EPID images using projections through a phantom to detect gantry angle per image and (3) comparison of planned and measured treatment delivery parameters. The proposed apparatus used for QA delivery is illustrated in figure 1.
Figure 1. Illustration of apparatus used for QA delivery.
Download figure:
Standard imageThe purpose of modifying the clinical plan for QA is that the MV beam projects through a phantom at all control points (CPs) of the arc. This is accomplished by setting the collimator to zero and opening the jaws and a pair of inferior MLCs accordingly. A CP from an original VMAT plan and a QA plan with such a modification are shown in figure 2.
Figure 2. Original clinical VMAT arc (left) and modification for QA delivery (right); the couch and collimator angles are set to zero; the collimator jaws and two most inferior leaf pairs are opened for all CPs.
Download figure:
Standard imageThe purpose of the phantom is to determine the gantry angle at which each projection image is acquired. The choice of phantom should be such that (1) a robust and accurate detection of gantry angle can be performed, and (2) it should minimize scatter from the MV beam incident on the phantom that reaches the area of the detector used to verify the clinical plan. In reality, EPID images are acquired over a finite timeframe that corresponds to a gantry rotation rather than a single gantry angle. Furthermore, while some proposed EPID-based VMAT techniques utilize a high imaging frequency with negligible gantry rotation per image (Bakhtiari et al 2011, Mans et al 2010), other techniques utilize a slow image rate or series of integrated images (Nicolini et al 2008, Iori et al 2007, Van Esch et al 2004) resulting in considerable rotation of the gantry per image. Hence ideally the phantom should be able to determine not just the mean gantry angle per image, but also the gantry rotation through which each image is acquired. Finally, as the projection through which the gantry angle is measured is achieved by opening MLC leaves, it is also of interest that the phantom be thin so that the gantry angle can be detected using one to two MLC leaf pairs. Phantoms have been previously described that are capable of measuring gantry angle of projection images (Ansbacher et al 2010, Mao et al 2008), which may be applicable for this purpose.
Proposed EPID pre-treatment QA techniques can be classified into two camps: intensity-based verification (Mans et al 2010, Nicolini et al 2008, Iori et al 2007, Van Esch et al 2004), and detection and verification of MLC positions (Bakhtiari et al 2011). For an intensity-based comparison, a predicted image is derived from plan parameters and compared to the measurement for the portion of the VMAT arc through which the image is acquired. In these cases the measured gantry rotation per image can be used in combination with the plan to determine the expected intensity per image. An intensity-based comparison has also been proposed for multiple sub-arcs (Nicolini et al 2008); in this case, the expected rotation of the gantry through which each image is acquired is known and the measured gantry rotation per image can be compared directly to the expected value. For EPID-based detection and verification of MLC positions (Bakhtiari et al 2011), the gantry angle recorded in the header of the DICOM file by the vendor was used to calculate the expected MLC positions per image. In this case, the gantry angle measured using the phantom could be used instead of the vendor supplied gantry angle to calculate expected MLC positions per image.
2.2. Specific implementation
A flowchart illustrating the process we used for clinical application is shown in figure 3. To evaluate feasibility, we implemented the proposed technique utilizing a standard Varian EPID (a-Si flat panel, IAS3 Varian software, SID = 130 cm, 1024 × 768 pixels, pixel size = 0.392 × 0.392 mm2 at the detector plane) installed on a Novalis Tx with a high definition MLC. We modified clinical treatment plans for QA using software developed in house in MATLAB 7.0 (The MathWorks Inc., Natick, MA); the two most inferior MLC leaf pairs and the X-jaws were opened to ±7.5 cm, the lower Y-jaw was opened to expose the most inferior MLCs (10.9 cm), and the collimator and couch angles were set to zero. Prior to QA delivery, a custom-built phantom (coil phantom) was positioned so as to be visible in the beam's eye view of the opened MLCs, after which it was localized exactly using a set of setup images at known gantry angles. The coil phantom was designed to measure gantry angle and consists of five gold coils (Visicoil, 0.8 mm diameter, 3.5 cm length) embedded in styrofoam in the pattern illustrated in figure 4. The center coil is oriented near the axis of rotation, with the three outer coils oriented in a triangular pattern each at a distance of ∼5 cm from the axis of rotation. The fifth coil is 1.8 cm from the axis of rotation and is oriented so that for any given MV projection of the coils, no three overlap each other simultaneously. The coils are thin and high contrast resulting in a sharp, high contrast projection in MV radiographs, while the styrofoam provides stability and causes minimal attenuation or scattering of the beam. The phantom was suspended off the end of the treatment couch so that the couch is not in the beam during QA delivery. The phantom was oriented so that the long axis of the coils was parallel with the gantry rotational axis (illustrated in figures 2 and 5(a)), so that as the gantry rotated and projection images were acquired, a sinogram with a distinct track per coil was created when a profile was taken across the imager (illustrated in figure 5(b)).
Figure 3. Flowchart of EPID-based VMAT QA procedure.
Download figure:
Standard image
Figure 4. Custom-built phantom (coil phantom) used for measuring gantry angle, consisting of five gold coils (0.8 mm diameter) embedded in styrofoam.
Download figure:
Standard image
Figure 5. Detection of gantry angle. An MV projection of the coil phantom is acquired during finite rotation of the gantry (a). The positions of the coils within the phantom define a sinogram with varying gantry angle, the prediction for which is shown for 0° (b) and 5° (c) gantry rotation per acquisition. Also shown is the 1D profile for a selected gantry angle (d).
Download figure:
Standard imageThe imager can be operated in an 'integrated acquisition' mode in which a single image is acquired for the entire treatment field, and a 'continuous acquisition' (or 'cine') mode in which individual sequential image frames are acquired in real time throughout the treatment field. We utilized the 'cine' mode for QA delivery. This mode is synchronized with the linear accelerator beam pulses so that image lines are read out between pulses; because of this, the frame rate of the imager is proportional to the beam pulse rate (and hence dose rate). Prior to transfer to the imaging computer, an adjustable number of 'frames' is averaged together into a cine image, with a series of cine images being acquired per treatment field. The characteristics of this acquisition mode were investigated for VMAT QA by McCurdy and Greer (2009); their study found the dose linearity, reproducibility of response and image stability to be acceptable for VMAT QA applications except in the case of very low total MU (mainly due to the dropping of ∼1 acquired image at the end of each acquisition). We found that system stability issues arose for our specific linear accelerator and imaging system when more than ∼200 total cine images were acquired during an arc delivery, in which the images failed to transfer to the imaging computer. Therefore for our evaluation, the number of frames averaged per image was adjusted to give an acceptable number of cine images per VMAT acquisition.
To measure the gantry rotation per acquired image, we compared a projection of the coil phantom in the open leaf pair of the EPID image to a template projection created analytically. For a projection where no gantry rotation occurs, the template projection is constructed using the analytical calculation of attenuation through cylindrical coils. The template for a projection that includes gantry rotation is constructed by taking the average (weighted by planned MU per gantry angle) of projections at static gantry angles every 0.1° over the angular range of the gantry rotation. Figure 5 illustrates the construction of the analytical projection of the coils for one line across the imager (orthogonal to the axis of gantry rotation and the long dimension of the coils). Attenuation of the beam by the coils is calculated for the projection image at the specified gantry angle (figure 5(a)), where the coils' locations and characteristics are pre-determined from the setup images. Figure 5(b) shows the sinogram of a 1D analytical projection of the coils when no gantry rotation occurs during image acquisition. Figure 5(c) shows the change in the sinogram when a 5° gantry rotation occurs during each image acquisition; for this case some blurring occurs leading to variations in peak width and intensity. Blurring is the greatest for the coils labeled 4 and 5 in figure 5, because they have the largest motion in projection space when the MV source is at −45°. Figure 5(d) shows the 1D analytical projection at a single gantry angle (mean gantry angle = −45°, indicated by the vertical lines in figures 5(b) and (c)) both when a 0° and 5° gantry rotation occurs during image acquisition. Finally, the background across each profile (leaf motion axis) of the measured projection was subtracted using a least-squares fifth-order polynomial fit, and the gantry rotation per image was determined by searching for the analytical projection of the coils that best matched the projection of the coils in the image.
2.3. Feasibility tests: gantry angle detection, phantom scatter, imager noise and linearity
We performed a number of investigations to evaluate the feasibility and accuracy of incorporating gantry angle verification into EPID-based VMAT QA. The aspects of the procedure that were investigated included: (1) accuracy of measuring mean gantry angle per cine image, (2) accuracy of measuring the angle through which the gantry rotates per cine image, (3) quantifying scatter incident on the detector due to the presence of the coil phantom and (4) evaluating noise and linearity of the portal imager pixel intensity for the cine acquisition mode.
We simultaneously evaluated the accuracy to which mean gantry angle and gantry rotation per image could be measured using the coil phantom. We first evaluated these using the integrated acquisition mode with the phantom aligned by CBCT for eight integrated MV images acquired at gantry angles that were equispaced so as to sample evenly throughout the projection space. For each of the eight gantry angles, integrated MV images were acquired with static gantry, as well as over arcs of 1°, 2°, 3°, 5° and 7°. Accuracy of measuring gantry rotation using the coil phantom was also evaluated for portal images acquired in the cine mode during delivery of a 360° arc with a total of 200 MU and uniformly planned MU per gantry angle. The delivery was performed with the number of frames averaged varying from 1 to 8, resulting in 1.4°–10.9° gantry rotation per image and 33–211 images acquired.
The scatter incident on the detector due to the presence of the coil phantom was quantified with the gantry at 0°. Integrated images were acquired with and without the coil phantom in place, and the difference between the two images was assumed to be due to the scatter from the phantom. The scatter profile was determined both for an open beam and for a beam with the jaws collimated down to the area used to image the phantom.
The EPID 'cine mode' is synchronized with the beam pulses so that the image lines are read out between pulses; often the entire image must be read over multiple pulses. The cine mode may be subject to potential artifacts such as imperfect synchronization, lost pulses, etc. Therefore it is important to quantify image noise and pixel intensity linearity with delivered MU. We evaluated the pixel intensity noise of the EPID operated in the cine mode across each image (intra-image noise) and over the series of acquired images (inter-image noise) for two types of VMAT plans, each consisting of a 360° arc with an open field. The CPs for the two plans each had a unique pattern for planned MU per gantry angle: (1) uniform MU/gantry angle (total MU = 200), and (2) linearly increasing MU per gantry angle. Two plans with linearly increasing MU per gantry angle were used; these included a ∼10.5-fold total increase in MU/degree from start to finish (MU between first two CPs = 0.2, MU between last two CPs = 2.0, total MU = 196), and a ∼64-fold increase in MU/degree (MU between first two CPs = 0.2, MU between last two CPs = 13.9, total MU = 1241). To evaluate intra-image noise, the pixel intensity across each image was compared to the mean pixel intensity over all images for a region of interest (ROI) defined as the portion of the image that had an open field but excluding the projection through the coil phantom (12 × 17 cm2 at isocenter, 398 × 564 pixels). To evaluate inter-image noise and pixel linearity, the mean pixel intensity of an ROI at the central axis (4 × 4 cm2 at isocenter, 130 × 130 pixels) was recorded over all images acquired for each plan and compared to the expected delivered MU/image.
2.4. Feasibility tests: QA of clinical VMAT plans
We performed an intensity-based verification of the treatment plan parameters of ten VMAT fields taken from clinical treatment plans. We calculated the predicted integrated image between each pair of consecutive CPs, accounting for the modification for QA delivery. A number of methods have been described to predict integrated portal imager response for dynamic MLC (DMLC) motion (Van Esch et al 2004, Warkentin et al 2003, Rosca and Zygmanski 2008). In our case, the commercial treatment planning system (Eclipse version 10.0, Varian Medical Systems Inc., Palo Alto, CA) has a clinically implemented function to calculate expected portal images for IMRT that we utilized. In this implementation, fluence is derived from the DMLC plan after which it is convolved with Gaussian kernels to account for scatter. We resampled the EPID images to match the prediction using linear interpolation between the cumulative sum of acquired intensity values per pixel, including a measurement and correction for EPID sag per image.
QA analysis consisted of 2D γ-index comparisons (Low et al 1998) of the projection of measured and predicted EPID response for summations through each axis: integrated cine images over the entire VMAT arc, sum of all cross-plane EPID profiles as a function of gantry angle and sum of all in-plane EPID profiles as a function of gantry angle. The EPID QA analysis was repeated using the gantry angle recorded in the DICOM header of each cine image as opposed to the gantry angle measured using the coil phantom.
3. Results
The left side of table 1 shows the error in measuring mean gantry angle and rotation per portal image for integrated acquisitions (eight equispaced images per set arc length). The '0° arc length' is the acquisition with a static gantry. Over all integrated acquisitions, the gantry angle was measured to within 0.0 ± 0.3°, and the rotation of the gantry per image to within 0.2 ± 0.2°. The results for cine mode acquisitions are shown on the right side of table 1 (360° rotation, 200 MU per acquisition). Specifically, given are the nominal, measured and calculated gantry rotations per cine image. The nominal gantry rotation is defined as the quotient of the total VMAT arc length by the total number of cine images. Measured gantry rotation refers to the mean (± standard deviation) gantry rotation per image measured using the coil phantom. The calculated gantry rotation is the mean (± standard deviation) difference between gantry angles of consecutive images, where the gantry angle per image was the mean gantry angle measured with the coil phantom or the gantry angle recorded in each cine image DICOM header. It should be noted that for the calculated gantry rotation, the mean value by definition falls near the nominal value, while the standard deviation indicates degree of variation from uniformly distributed gantry angles.
Table 1. Error (mean ± standard deviation) in measuring gantry angle and rotation per cine image using the coil phantom for acquisitions in integrated and cine modes.
| Integrated acquisition | Gantry rotation per cine image | |||||
|---|---|---|---|---|---|---|
| Measurement error | ||||||
| Set arc length | Mean angle | Rotation | Nominal | Measured | Calculated (Phantom) | Calculated (DICOM) |
| 0° | 0.0 ± 0.2° | 0.8 ± 0.3° | 1.4° | 1.5 ± 0.2° | 1.4 ± 0.2° | 1.4 ± 1.2° |
| 1° | 0.0 ± 0.2° | 0.2 ± 0.3° | 2.8° | 2.9 ± 0.3° | 2.8 ± 0.3° | 2.8 ± 1.3° |
| 2° | 0.1 ± 0.3° | 0.2 ± 0.3° | 4.2° | 4.3 ± 0.3° | 4.2 ± 0.2° | 4.2 ± 0.9° |
| 3° | 0.1 ± 0.3° | 0.1 ± 0.2° | 5.5° | 5.7 ± 0.4° | 5.6 ± 0.3° | 5.6 ± 1.5° |
| 5° | 0.1 ± 0.4° | 0.0 ± 0.3° | 10.9° | 11.1 ± 0.8° | 11.2 ± 0.6° | 11.2 ± 1.2° |
| 7° | 0.1 ± 0.4° | −0.2 ± 0.2° | ||||
| All arcs | 0.0 ± 0.3° | 0.2 ± 0.2° | ||||
Figure 6 shows the histograms of the difference from the nominal distance between portal images (1.4°) for that measured using the coil phantom (a), calculated as the difference between the mean gantry angles (measured with the coil phantom) of consecutive cine images (b), and calculated as the difference between the gantry angles recorded in the DICOM header of consecutive cine images (c). The spread of the spectrum in figure 6(c) confirms previous studies (Ansbacher et al 2010), indicating that the vendor-provided gantry angle in the DICOM header lacks the necessary precision to verify gantry angle during VMAT delivery, and shows that an independent technique to verify the gantry angle is warranted.
Figure 6. Histograms of difference from expected arc length per image (1.4°) when the arc length was measured using the coil phantom (a), calculated using the mean gantry angle from the coil phantom (b), and calculated using the angle in the DICOM image header (c).
Download figure:
Standard imageThe scatter incident on the detector due to the presence of the coil phantom was found to be negligible both with an open and closed field (<<1%).
For the VMAT arc with an open field and uniform MU per degree, the standard deviation of the intra-image noise ranged from 1.5% of the image signal minus background (5.6 MU and 5.0° per image) to 5.6% (1.4 MU and 1.25° per image). For the VMAT arc with an open field and a linear increase in MU per degree, the pixel intensity was highly linear. For the ∼10.5-fold and ∼64-fold increase in MU per degree, the correlation coefficient between prediction and expected was 0.993 and 0.999, respectively, and the standard deviation of the difference from expected was 2.8% and 1.5% of the maximum signal, respectively.
A summary of the results for the QA procedure carried out for ten clinical VMAT arcs is shown in table 2. A relative gamma analysis (Low et al 1998) (3% of global maximum, 3 mm, 3°, 10% threshold) was used for comparison between predicted and measured EPID response for the projection of the data through each axis. The resulting projections through each axis for arc number 7 in table 2 are shown in figure 7. In figure 7, the left column represents integrated cine images over the entire VMAT arc, the middle column represents the sum of all cross-plane EPID profiles as a function of gantry angle and the right column represents the sum of all in-plane EPID profiles as a function of gantry angle. Using the DICOM gantry angle rather than the gantry angle measured by the coil phantom resulted in decreased agreement between measured and predicted for cross leaf versus gantry and leaf motion versus gantry planes (p < 0.001).
Figure 7. Summation through each axis for acquired and predicted EPID response for arc 7 of table 2. Also shown is the absolute value of the difference between measured and expected.
Download figure:
Standard imageTable 2. Parameters corresponding to ten clinical VMAT arcs for which the EPID QA technique was performed. Dose rate modulation factor is the ratio of the maximum to minimum planned MU between any two CPs.
| Plan parameters | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Plan no. | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | |
| Treatment site | Brain | Brain | Brain | Brain | Brain | Brain | Brain | Prostate | Prostate | Spinal | |
| and nodes | and nodes | cord | |||||||||
| No. of PTVs | 1 | 1 | 1 | 1 | 2 | 2 | 3 | 1 | 1 | 1 | |
| MU | 884 | 981 | 1138 | 3173 | 432 | 1350 | 91 | 184 | 251 | 437 | |
| Field size | 5.4 × 4.9 | 7.6 × 7.6 | 6.3 × 5.2 | 5.0 × 4.7 | 7.1 × 7.3 | 16.0 × 14.4 | 16.9 × 8.7 | 18.8 × 19.2 | 11.1 × 11.1 | 10.8 × 11.5 | |
| Arc length (deg) | 358 | 358 | 290 | 358 | 160 | 358 | 170 | 360 | 360 | 360 | |
| No. of CPs | 178 | 178 | 178 | 178 | 98 | 178 | 97 | 147 | 177 | 177 | |
| Modulation factor | 2.3 | 2.7 | 1.4 | 1.2 | 1.8 | 1.8 | 4.2 | 1.9 | 1.4 | 3.7 | |
| QA parameters | |||||||||||
| No. of images acquired | 176 | 195 | 180 | 151 | 172 | 211 | 139 | 147 | 195 | 187 | |
| Imaging angular resolution (deg) | 2.0 | 1.8 | 1.6 | 2.4 | 0.9 | 1.7 | 1.2 | 2.4 | 1.8 | 1.9 | |
| QA results (using phantom) | γ index≤1, 3% global maximum, 3 mm, 3° | Overall | |||||||||
| Cross leaf versus leaf motion | 100.0 | 99.8 | 99.8 | 100.0 | 99.2 | 96.5 | 96.1 | 96.9 | 99.1 | 99.2 | 98.7 ± 1.5 |
| Cross leaf versus gantry | 95.0 | 95.5 | 95.7 | 97.7 | 95.1 | 94.8 | 86.8 | 83.5 | 89.2 | 92.9 | 92.6 ± 4.6 |
| Leaf motion versus gantry | 95.2 | 96.1 | 96.1 | 98.8 | 94.3 | 94.4 | 85.2 | 71.4 | 85.1 | 91.4 | 90.8 ± 8.2 |
| QA results (using DICOM) | γ index ≤ 1, 3% global maximum, 3 mm, 3° | Overall | |||||||||
| Cross leaf versus leaf motion | 100.0 | 99.9 | 99.8 | 100.0 | 99.2 | 96.5 | 96.4 | 96.5 | 99.1 | 99.3 | 98.7 ± 1.6 |
| Cross leaf versus gantry | 83.3 | 79.6 | 88.1 | 96.9 | 78.3 | 85.9 | 56.0 | 68.3 | 71.0 | 71.8 | 77.9 ± 11.6 |
| Leaf motion versus gantry | 72.2 | 71.6 | 87.9 | 97.1 | 64.5 | 78.3 | 40.5 | 47.1 | 57.3 | 59.4 | 67.6 ± 17.5 |
In table 2, the majority of the QAs of clinical VMAT plans had γ-index passing rates above 90% with the exception of the cross leaf versus gantry and leaf motion versus gantry axes for plans 7–9. Upon closer examination it was found that the passing rates in these axes began to decrease when the MU per image fell below ∼5, which, along with the evaluation of imager noise, indicates that noise increases as MU per image decreases for the continuous acquisition mode. The fields with the lowest passing rates (fields 7–9 in table 2) had MU per image below 2. This has implications for pixel-intensity-based pre-treatment QA techniques utilizing cine imaging. One management strategy for this limitation is to decrease the cine acquisition frame rate until the MU per image reaches an acceptable level, or it could be avoided altogether by using one of the alternative EPID QA methods such as detection and verification of MLC positions (Bakhtiari et al 2011). Table 2 also indicates that a summation of cine images into a single integrated image (cross leaf versus leaf motion) resulted in a high degree of agreement with the predicted results regardless of frame rate.
4. Discussion
We have proposed a method to add gantry angle to the group of treatment plan parameters that can be verified by EPID-based pre-treatment QA by essentially encoding gantry angle into the image itself. One disadvantage is that the method presented here requires some modifications to the original treatment plan, namely changing jaw positions, setting collimator to zero and opening the most inferior leaf pairs. While ideally the QA plan should be unchanged from the original plan, for practicality some changes are often accepted for various QA techniques (such as setting gantry and collimator angles to zero for all IMRT beams, or splitting of a VMAT arc into multiple arcs). For our case one option to avoid plan modifications would be to place the coil phantom directly in the unmodified field; however, the movement of the leaves would result in obscurity of part of the phantom and the phantom would attenuate the primary fluence and have a greater effect on the scatter. This would require a more complex gantry angle measurement technique and implementation of a sophisticated correction for the presence of the coil phantom. A more plausible option is to perform the QA with and without modifying the plan, with gantry angle being monitored only for the delivery of the modified plan. This may also be a useful option in cases where the most inferior leaves that are opened to image the phantom are also utilized in the clinical VMAT plan. In addition to pre-treatment QA, there is a potential to apply the EPID-based gantry measurement to VMAT commissioning and routine machine QA, where there are no such concerns about the modification of a clinical plan.
One observation from table 1 and figure 6 is that the measured and calculated gantry rotation from the coil phantom results in distributions that are tightly grouped about the expected value (a and b), whereas the distribution from the DICOM header angle is multimodal. This is most likely due to the DICOM header information not being updated very frequently and is in agreement with observations from a previous study which found the DICOM angle to have limited precision (Ansbacher et al 2010). The limited precision of the gantry angle recorded in the DICOM file resulted in decreased agreement with prediction and is a strong argument in favor of an independent verification of EPID gantry angle for VMAT QA. The coil phantom was developed in house and was sufficient for this feasibility study; other phantoms have been suggested that could be incorporated and could potentially offer a more accurate measurement (Ansbacher et al 2010, Mao et al 2008).
For our specific setup we had system stability issues when ∼200+ images were acquired during the QA. The angular resolution achievable when 200 images are acquired is at minimum 1.8° per image; this should be sufficient for most cases since it often allows for better than one image per CP. Furthermore, this limit is specific to our apparatus and was not an issue for some reported EPID QA techniques (Bakhtiari et al 2011).
In this study we used a linear interpolation of the measured image data to match the prediction. We may expect some interpolation error, which will be greatest when the angular distance between images is much larger than the angular distance between CPs. However, this error should be small for the clinical cases presented here because the angular resolution of the images was comparable to that of the CPs. One option to minimize interpolation error if a lower imaging angular resolution is used would be to interpolate the dataset with the highest angular resolution down to the lower resolution dataset, rather than always interpolating the measurement to the prediction. We found pixel linearity to be valid for VMAT arcs with a ∼10.5- and ∼64-fold change in MU per CP. This range tested is well above any of the dose rate modulations observed in clinical plans; the dose rate modulation for arcs tested in table 2 ranged from 1.2 to 4.2.
5. Conclusion
An independent measurement of the gantry angle per image has been incorporated into current methods of EPID-based pre-treatment QA for VMAT, adding gantry angle to the treatment plan parameters that can be verified using an EPID. This technique can be a valuable and efficient alternative to other VMAT QA systems, and has potential for future applications in VMAT commissioning and routine machine quality assurance.
Acknowledgment
We thank Fang-fang Yin, Mike Scribner and Jim Adams for helpful comments and discussions.