Assessment of error in the MV radiation isocenter position calculated with the Elekta XVI software

Abstract The assessment of the coincidence of imaging and radiation isocenters is an important task of regular quality assurance of medical linear accelerators (linacs) as recommended in national and international quality assurance guidelines. A previously reported investigation of the accuracy of the Elekta XVI software to localize the linac radiation isocenter, by comparing statistically with other independent software, has shown some discrepancies at the sub‐mm level. A further investigation is carried out here using a set of reference images and mathematical operations to observe how the Elekta XVI software analyses them. Symmetric mathematical operations on reference images should result in symmetrical outcomes. Three different rotation functions are used in increasing degree of complexity to characterize the Elekta XVI software error in the linac radiation isocenter position. No independent algorithms or phantoms are used in this methodology. The magnitude and direction of the radiation isocenter localization error has been determined to be consistently 0.13 mm or 0.14 mm in the longitudinal direction towards the target depending on the case. The radiation isocenter localization error comprises two separated errors of the Ball Bearing Center by 0.13 mm and MV Field Center by either 0.00 mm or −0.01 mm in the longitudinal direction towards the target. The calculation of the MV Field Center is influenced by the polymethyl methacrylate rod supporting the ball‐bearing. The precise value and the root cause of the error cannot be assessed due to the rounding effect of the results reported by the Elekta XVI software and lack of access to the source code.

is to independently verify the coincidence of the imaging and radiation isocenters, as recommended in quality assurance guidelines 1 Riis et al 2 reported that the linac radiation isocenter position determined using the Elekta X-ray Volume Imaging (XVI) software differs to their independent method, by an average of 0.23 ± 0.13 mm in the longitudinal direction towards the target. This error adds uncertainty to the overall patient setup uncertainty, including registration and table top movement. All such uncertainties should be minimized; however where a systematic error can be identified, it should be corrected. No clear explanation of the cause of this discrepancy was provided, but Riis et al suggested that it might be due to an issue with the Elekta XVI software. In this study, further investigation is undertaken to assess the accuracy of the radiation isocenter position calculated by the Elekta XVI software.

2.B | Radiation isocenter localization procedure
The radiation isocenter localization procedure is part of the kilovoltage (kV) imager flexmap calibration process as described in the Elekta XVI corrective maintenance manual. 3,4 The kV flexmap calibration process aims to correct clinical kV images for the sag of both the kV EPID panel and kV source due to gravity while the gantry rotates. The correction is stored in the form of a two-dimensional table (called kV flexmap) mapping the shift that needs to be applied to the kV image in two perpendicular directions at any gantry angle.
The kV flexmap calibration process attaches a ball-bearing phantom to the patient support table and the ball-bearing is set up to be close to the linac isocenter. Eight megavoltage (MV) images of the ballbearing are acquired in total, two at each of the four cardinal gantry angles, using two diametrically opposite collimator angles. The images are analyzed in the Elekta XVI software which calculates the shifts required to be applied to the ball-bearing position to place it at the linac radiation isocenter. This corrected position is used to further create kV flexmaps for the kV imager panel to ensure the alignment of the kV imaging and MV treatment isocenters.

2.C | Error estimation methodology
Testing software based on expected output versus controlled input is considered a definitive technique of software validation. To control the input to the Elekta XVI software, simple mathematical operations were applied to an original set of input images.  F I G . 1. One example of eight original images of the reference ballbearing used in the Elekta XVI software.

2.C.2 | Method 2
The second method investigated if the polymethyl methacrylate (PMMA) rod supporting the ball-bearing affects the processing of the images in the Elekta XVI software. Two sets of images were created with directions reversed to each other and with the PMMA rod in the second set oriented perpendicular to the original set, i.e. one set was created by applying a rotation of −90°and the second set with a rotation of +90°.

2.C.3 | Method 3
The third method investigated residual shifts of the ball-bearing from symmetric images. It is expected that symmetric images processed by the Elekta XVI software should result in symmetric shifts of the ball-bearing, i.e. zero residual shift in each direction.

2.C.4 | Dataset dependency
To check if the results depend on the input dataset, six different input sets of images from different days were considered (five sets using the 6 MV beam and one set using the 15 MV beam).

2.C.5 | Linac dependency
To check if the results depend on the specific linac, two additional tests were performed: a input images from the Elekta Versa HD TM linac were analyzed on a different linac namely an Elekta Synergy® with the same Elekta iViewGT and XVI software version.
b input images from the Elekta Synergy® linac were analyzed on both linacs

| RESULTS AND DISCUSSION
The results from method 1 showed that the Elekta XVI software incorrectly calculated the shift of the reference ball-bearing to the radiation isocenter by 0.13 mm in the longitudinal direction towards the target. Zero shifts were calculated in the lateral and vertical directions (see Table 1). This outcome confirms the Riis et al suspicion that the Elekta XVI software might cause some discrepancy in the radiation isocenter localization.
The results from method 2 showed that the Elekta XVI software calculated the shift of the reference ball-bearing to the radiation isocenter by 0.135 mm in the longitudinal direction towards the target, a slightly higher value, by 0.005 mm, than observed from method 1. Also the shift in the vertical direction was noted to be changed from zero to 0.005 mm downwards. No change was observed in the lateral direction (see Table 2). The variation of results due to the direction of the PMMA rod is negligible, but nevertheless non-zero. To further confirm if this error was reproducible another dataset was used and the dependency on the rod direction was not observed. This indicates that results depend, at this small level, on how the Elekta XVI software processes a specific dataset. This is in agreement with Riis et al's findings that "the phantom asymmetry does not appear to cause the discrepancy". The possible issue might be due to the sensitivity of the software to the quality of the input images such as noise and contrast, however given the level of effect, this was not further investigated.

Method 3 created a symmetric set of images that should result
in the Elekta XVI software calculating no shifts in any directions; however the Elekta XVI software calculated the shift of the reference ball-bearing to the radiation isocenter by 0.14 mm in the longitudinal direction towards the target. Zero shifts were calculated in the lateral and vertical directions (see Table 3). This outcome differs by 0.005-0.01 mm from the results of methods 1 and 2.    Fig. 3, where the extensive blurring effect of the ball-bearing is clearly visible.

Direction
The linac dependency test, also using the method 3 algorithm, showed that images analyzed on a different linac (Elekta Synergy®) give exactly the same results for one particular set of original images  Table 4). This  Table 4). The absolute value of the MV

| CONCLUSION
It has been determined that the Elekta XVI software used to define kV flexmaps for the image guidance system incorrectly calculates the

CONF LICT OF I NTEREST
The authors declare no conflict of interest.