EXACTRAC

x-ray and beam isocentersWhats the difference?

Dennis Tideman Arpa) and Jesper Carl

Department of Medical Physics, Oncology, Aalborg Hospital, Aarhus University Hospital, Hobrovej 18-22, DK-9100 Aalborg, Denmark

(Received 7 August 2011; revised 30 January 2012; accepted for publication 30 January 2012; published 23 February 2012) Purpose: To evaluate the geometric accuracy of the isocenter of an image-guidance system, as implemented in the EXACTRAC system from BRAINLAB, relative to the linear accelerator radiation isocenter. Subsequently to correct the x-ray isocenter of the EXACTRAC system for any geometric discrepancies between the two isocenters. Methods: Five Varian linear accelerators all equipped with electronic imaging devices and EXACTRAC with robotics from BRAINLAB were evaluated. A commercially available Winston-Lutz phantom and an in-house made adjustable base were used in the setup. The electronic portal imaging device of the linear accelerators was used to acquire MV-images at various gantry angles. Stereoscopic pairs of x-ray images were acquired using the EXACTRAC system. The deviation between the position of the external laser isocenter and the EXACTRAC isocenter was evaluated using the commercial software of the EXACTRAC system. In-house produced software was used to analyze the MV-images and evaluate the deviation between the external laser isocenter and the radiation isocenter of the linear accelerator. Subsequently, the deviation between the radiation isocenter and the isocenter of the EXACTRAC system was calculated. A new method of calibrating the isocenter of the EXACTRAC system was applied to reduce the deviations between the radiation isocenter and the EXACTRAC isocenter. Results: To evaluate the geometric accuracy a 3D deviation vector was calculated for each relative isocenter position. The 3D deviation between the external laser isocenter and the isocenter of the EXACTRAC system varied from 0.21 to 0.42 mm. The 3D deviation between the external laser isocenter and the linac radiation isocenter ranged from 0.37 to 0.83 mm. The 3D deviation between the radiation isocenter and the isocenter of the EXACTRAC system ranged from 0.31 to 1.07 mm. Using the new method of calibrating the EXACTRAC isocenter the 3D deviation of one linac was reduced from 0.90 to 0.23 mm. The results were complicated due to routine maintenance of the linac, including laser calibration. It was necessary to repeat the measurements in order to perform the calibration of the EXACTRAC isocenter. Conclusions: The deviations between the linac radiation isocenter and the EXACTRAC isocenter were of an order that may have clinical relevance. An alternative method of calibrating the isocenter of C the EXACTRAC system was applied and reduced the deviations between the two isocenters. V 2012 American Association of Physicists in Medicine. [DOI: 10.1118/1.3685581] Key words: image guided, Winston Lutz test, isocenter

I. INTRODUCTION The introduction of modern image guided radiotherapy (IGRT) technology has improved precision and accuracy in patient setup before administration of radiotherapy. This again has allowed for the use of decreasing clinical target volume (CTV) to planning target volume (PTV) margins. Reduction in CTV-PTV margins and volume of normal tissue implies a reduction in both acute and late toxicity. This has been demonstrated in radiotherapy of localized prostate cancer.1,2 A geometric accuracy of up to 2 mm has been reported for patient setup using image-guidance based on stereoscopic pairs of x-rays as implemented in the V 3,4 EXACTRAC (ET) system from BRAINLAB. The ET system makes use of an external laser system to represent the actual radiation isocenter of the linear accelerator (linac). But this method involves a human observer to perform the phantom
R

alignment with the external lasers. Also the external lasers may be exposed to, e.g., drift or mechanical inuences, so the actual accuracy may in fact be degraded5 and systematic shifts introduced.6 Systematic errors of 1 mm in setup may have clinical relevance, as one study demonstrated that reduction of CTV-PTV margin to only 3 mm, may have caused a negative inuence in terms of an increased number of patients with biochemical failure following radiotherapy of localized prostate cancer.7 A Winston-Lutz (WL) test may be used to measure the accuracy of the linac radiation isocenter. The test uses a phantom with a small embedded metal sphere combined with portal images taken from various gantry angles. Combined with digital detection of external lasers this may allow discrepancies between laser and radiation isocenter to be less than 0.25 mm.8 Another approach would be to eliminate the use of external lasers and use the WL phantom to align the x-ray isocenter of the ET
C V 2012 Am. Assoc. Phys. Med.

1418

Med. Phys. 39 (3), March 2012

0094-2405/2012/39(3)/1418/6/$30.00

1418

1419

D. T. Arp and J. Carl: X-ray and linac isocenters

1419

system with the linac radiation isocenter. The aim of the present study was to design a WL based method to detect the mean position of the linac radiation isocenter (average isocenter) to evenly distribute the effect of gantry sag. Subsequently the aim was to correct the x-ray isocenter of the ET system for the linac radiation isocenter shift. II. MATERIALS AND METHODS
II.A. The accelerators

Five Varian medical linacs (two 2100 C/D and three iX) equipped with electronic portal imaging device (EPID) of type as500 or as1000. All linacs were equipped with ET including Robotics from BRAINLAB, which enables patient setup in six degrees of freedom (three axis of translation and rotation of the couch). The linacs were energy matched with identical beam-congurations, making it possible to treat the same patient independently of the choice of linac.
II.B. The WL phantom
FIG. 1. WL phantom attached to adjustable base on the linac treatment table and positioned in laser isocenter.

The WL phantom consisted of a commercially available (Varian) polyester 5 cm cube with a 2 mm embedded steel

FIG. 2. Binary representation of the 4 4 eld (a) and a cross-correlated image (b) constructed between center area of DICOM image and a template (c). Sphere and radiation isocenter detection using the detection algorithm (d). The circle represents the sphere detection and the cross the beam axis center. A Gaussian t to the cross-correlated image around center pixel in the Lng (Gaussian t 1) and Vrt (Gaussian t 2) direction (e). The pixel positions represent the position along the axis relative to the center pixel found as the brightest pixel. Medical Physics, Vol. 39, No. 3, March 2012

1420

D. T. Arp and J. Carl: X-ray and linac isocenters

1420

FIG. 3. The detected deviations of the radiation isocenter (R-L dots) and the ET system isocenter (X-L squares) in Vrt-Lng (a) and Lat-Lng directions (b). The lines connect the corresponding R-L and X-L deviations and represent the R-X deviations. The error bars represent the standard deviation of the averaged isocenter deviation from linac 1.

sphere in the center and external markings (black lines). The phantom was supplied with an in-house made PMMA adjustable base that allows for translation of the phantom in 0.1 mm steps in all three axes Lateral (Lat), Vertical (Vrt), and Longitudinal (Lng) direction (Fig. 1). The base was rmly attached to the treatment table using a bar xed to the edges of the table.
II.C. Setup procedure and image acquisition

were obtained to determine the deviation between the isocenter of the ET system and the laser isocenter (X-L). Subsequently, deviation between the radiation isocenter of the linac and the isocenter of the ET system was calculated (R-X). Using the adjustable base, the sphere was positioned in the average radiation isocenter of the linac within 0.1 mm. The isocenter of the ET system was then calibrated in this position and the R-X deviations evaluated.
II.E. The image analysis software

Using the adjustable base, the external markings on the WL phantom were aligned with the external laser system (laser isocenter) to measure the deviation between the linac radiation isocenter and the laser isocenter (R-L). A 4 4 cm eld was dened by the multileaf collimator (MLC) and portal images of the phantom were obtained for gantry angles 0 , 90 , 180 , and 270 . The portal imager was positioned at 140 cm from the radiation focus providing a spatial resolution in the isocenter of 0.56 and 0.28 mm for the as500 and as1000, respectively. The portal images were acquired in DICOM format using Varian AM Maintenance (AM Maintenance is a package for managing the EPID). The images were integrated over 100 MU to obtain a sufcient signalnoise ratio giving a well differentiated sphere edge denition. Leaving the phantom untouched in the same position, the WL mode in the ET system was used and x-ray images

Using a method similar to others,9 a MATLABV (The MathWorks, Inc.) script was produced to automatically analyze the DICOM images. The method used in this study applied a threshold leveling technique creating a binary representation of the 4 4 cm eld [Fig. 2(a)] and nding the center of the eld as the center-of-mass of the binary representation. This method ensures that the center of the eld is found independently of the effects of EPID sag. In the DICOM image a small, approximately 1 1 cm, area was cropped around the center of the eld, as found in the binary representation. The sphere was then detected using a cross-correlation between the image and a constructed template [Fig. 2(b)]. The template consisted of a 7 7 pixel 2D representation of the sphere, with the outline homogeneously valued as the
R

TABLE I. 3D deviation vector for linac and ET with corresponding linac and detector type and resolution for EPID and ET. R-L dev. r3Ddev (mm) 0.37 0.50 0.52 0.74 0.83 X-L dev. r3Ddev (mm) 0.21 0.42 0.42 0.30 0.41 R-X dev. r3Ddev (mm) 0.55 0.86 0.31 0.81 1.07

Linac nr. 1 2 3 4 5

Linac type Varian iX Varian iX Varian iX Varian2100 C/D Varian2100 C/D

EPID type as1000 as1000 as1000 as500 as500

EPID detector resolution(px) 1024 768 1024 768 1024 768 512 384a 512 384a Pixel size 0,28 mm

EXACTRAC

detector resolution(px)

512 512 512 512 512 512 512 512 512 512 Pixel size 0,26 mm

The lower resolution in EPID type as500 is scaled by bicubic interpolation to have same pixel size as as1000.

Medical Physics, Vol. 39, No. 3, March 2012

1421

D. T. Arp and J. Carl: X-ray and linac isocenters

1421

direction [Fig. 2(e)]. Using a method similar to others6,8,10 the average position of the linac radiation isocenter was determined and the deviation relative to the corresponding sphere position calculated. To ensure reproducibility of the setup and the detection algorithm several image sets were acquired with the WL phantom setup being fully repositioned in between acquisitions. Furthermore, the detection algorithm was veried by making controlled movements of the WL phantom in the Vrt, Lng, and Lat direction and comparing these to the detected displacements of the sphere with the detection algorithm. Analogously, the controlled displacements were compared to detected displacements using the ET system. III. RESULTS Deviations relative to the laser isocenter were calculated for ve linacs using the same setup, as described in the previous section. Three different deviations were evaluated the deviation between the linac radiation isocenter and the laser isocenter (R-L deviation), the deviation between the ET x-ray isocenter and the laser isocenter (X-L deviation), and the deviation between the linac radiation isocenter and the ET x-ray isocenter (R-X deviation).The R-L and X-L deviations are shown in Figs. 3(a) and 3(b), in a coordinate system with origin in the laser isocenter. To evaluate the reproducibility of the setup and the detection algorithm, four sphere detections of the same linac (linac 1) were preformed on different days over a period of one month. In Figs. 3(a) and 3(b), the linac 1 value is a mean of the four sphere detections and the error bars represent the standard deviation. The error bars include the uncertainties related to the setup and the reproducibility of both the detections algorithm and the ET system. The standard deviations are within 0.07 mm and are mainly due to the setup uncertainties. The reproducibility of the sphere detections by the detection algorithms and the ET software are within 0.01 mm. The length of the 3D deviation vector (3D dev.), dened by q 2 2 2 r3Ddev: rlat rvrt rlng ; with rlat , rvrt , and rlng being the lateral, vertical, and longitudinal R-L and X-L deviations are listed in Table I and ranges from 0.37 to 0.83 mm and 0.21 to 0.42 mm for the R-L and X-L deviations, respectively. Considering the R-X deviation the 3D dev. ranges from 0.31 to 1.07 mm. In Fig. 4, controlled displacements of the WL phantom and the corresponding sphere detection with the detection algorithm and the ET system are shown. Displacements of 0.125, 0.5, 1, and 2 mm in all three directions were performed using the adjustable base. These results verify the sphere detection algorithm within 0.1 mm of the intentional displacement and furthermore the submillimeter sensitivity of the adjustable base. A calibration of the ET isocenter to the average radiation isocenter was performed for one linac (linac 5). Table II shows the reduction of the R-X 3D dev. from 0.90 to 0.23 mm. It should be noted that these measurements were performed some time after the measurements shown in

FIG. 4. Controlled displacements of the WL phantom and the corresponding sphere detection with the detection algorithm and the ET system in Lat (a), Vrt (b), and Lng (c) direction, respectively.

brightest pixel in the cropped area. The outline of the sphere was surrounded with an averaged value of the background level [Fig. 2(c)]. Deviation between beam axis, represented as the center of the 4 4 cm eld, and the center of the sphere, representing the laser isocenter, was evaluated [Fig. 2(d)]. A subpixel accuracy was obtained using a Gaussian t to the cross-correlated image in both the Lng, Vrt, and Lat
Medical Physics, Vol. 39, No. 3, March 2012

1422

D. T. Arp and J. Carl: X-ray and linac isocenters

1422

TABLE II. Deviations before and after ET calibration for X-L, R-L and R-X deviation, respectively. Before calibration rlat (mm) X-L R-L R-X 0.09 0.15 0.24 rvrt (mm) 0.13 0.16 0.29 rlng (mm) 0.16 0.65 0.81 r3Ddev (mm) 0.22 0.69 0.90 rlat (mm) 0.01 0.18 0.17 After calibration rvrt (mm) 0.17 0.19 0.02 rlng (mm) 0.78 0.62 0.16 r3Ddev (mm) 0.80 0.67 0.23

Fig. 3 and Table I. The inconsistency between these deviations and the deviations in Table II are due to routine service maintenance of the linacs including laser calibration and ET calibration. Figure 5 shows the measured R-L and X-L deviations performed before laser calibration and the new measurements before and after calibration of the ET isocenter. IV. DISCUSSION In this study deviations, relative to the laser isocenter, for the linac radiation isocenter and the ET system isocenter were evaluated and a quick and precise linac isocenter detection method was developed using the EPID and an automatic detection algorithm. The results presented a maximum 3D dev. of 0.83 mm for the R-L deviation with a corresponding 3D dev. of 0.41 mm for the X-L deviation. This yielded a 3D dev. of 1.07 mm for the R-X deviation. It should be noted that this maximum value was measured on a linac equipped with an EPID of type as500, and the lower resolution of the detector could have overestimated the deviation. In our institution, a tolerance level of 0.70 mm deviation in Lat, Lng, and Vrt direction between the ET isocenter and the laser isocenter is used, in accordance with the accuracy of the ET system.4 Using the maximum tolerances in all three directions a 3D dev. of 1.2 mm is plausible. Considering the R-X 3D dev. ranged from 0.31 to 1.07 mm these deviations could have clinical relevance, especially in treatments having a clinical margin of 23 mm. Undoubtedly, the absolute R-L and X-L deviations are strongly related to the setup uncertainty of the WL phantom

with the external lasers and the laser alignment to the radiation isocenter. The relative deviation between the detected radiation isocenter and the ET isocenter was, however, independent of the external laser system and the inherent uncertainty. It was possible to adjust the sphere of the WL phantom with submillimeter accuracy, as shown in Fig. 4. Positioning the sphere in the averaged radiation isocenter and calibrating the ET isocenter to this position was done for one linac and showed a reduction of the R-X 3D dev. from 0.90 to 0.23 mm. Considering the stability of the deviations over time, this should allow for a reduction of the systematic deviations. Figure 5 illustrates not only the reduction of the systematic deviations, but also that these discrepancies are not reduced by the routine maintenance procedures. The calibration of the laser system and the following standard ET calibration only resulted in a reduction of the R-X 3D dev. from 1.07 to 0.90 mm. It also raises some issues regarding the uncertainty factors of the measurements, including interobserver variations. These effects might cause the error bars in Fig. 3 to be considerably larger. This study was performed on nonstereotactic treatment units being used for daily clinical treatment in our institution. The fact that the linacs are energy matched makes is possible to treat the same patient on different linacs, but in doing so the inherent geometric uncertainties of each linac must be taken into account, and therefore, a reduction of the interlinac differences could have clinical relevance. In stereotactic treatments, this is most certainly a clinically relevant subject and a continuous reduction of CTV-PTV margins only adds to its importance.

FIG. 5. R-L and X-L deviations before laser calibration and the deviations before and after calibration of the ET isocenter (both after laser calibration) in Vrt-Lng directions (a) and Lat-Lng directions (b). Medical Physics, Vol. 39, No. 3, March 2012

1423

D. T. Arp and J. Carl: X-ray and linac isocenters

1

1423 D. P. Dearnaley et al., Comparison of radiation side-effects of conformal and conventional radiotherapy in prostate cancer: A randomised trial, Lancet 353, 267272 (1999). 2 H. T. Chung et al., Does image-guided radiotherapy improve toxicity prole in whole pelvic-treated high-risk prostate cancer? Comparison between IG-IMRT and IMRT, Int. J. Radiat. Oncol., Biol., Phys. 73, 5360 (2009). 3 N. Hayashi et al., Assessment of spatial uncertainties in the radiotherapy process with the Novalis system, Int. J. Radiat. Oncol., Biol., Phys. 75, 549557 (2009). 4 D. Verellen et al., Quality assurance of a system for improved target localization and patient set-up that combines real-time infrared tracking and stereoscopic x-ray imaging, Radiother. Oncol. 67, 129141 (2003). 5 J. P. Vinci, K. R. Hogstrom, and D. W. Neck, Accuracy of cranial coplanar beam therapy using an oblique, stereoscopic x-ray image guidance system, Med. Phys. 35, 38093819 (2008). 6 J. Kim et al., Image-guided localization accuracy of stereoscopic planar and volumetric imaging methods for stereotactic radiation surgery and stereotactic body radiation therapy: A phantom study, Int. J. Radiat. Oncol., Biol., Phys. 79, 15881596 (2011). 7 B. Engels et al., Conformal arc radiotherapy for prostate cancer: increased biochemical failure in patients with distended rectum on the planning computed tomogram despite image guidance by implanted markers, Int. J. Radiat. Oncol., Biol., Phys. 74, 388391 (2009). 8 J. Grimm et al., A quality assurance method with submillimeter accuracy for stereotactic linear accelerators, J. Appl. Clin. Med. Phys. 12, 3365 (2011). 9 B. Winey, G. Sharp, and M. Bussiere, A fast double template convolution isocenter evaluation algorithm with subpixel accuracy, Med. Phys. 38, 223227 (2011). 10 Y. Miyabe et al., Positioning accuracy of a new image-guided radiotherapy system, Med.Phys. 38, 25352541 (2011).

Furthermore the R-L deviations were all averaged values of the Vrt, Lat, and Lng deviations and accepting the shift of the radiation isocenter when rotating the treatment head, as a consequence of, e.g., gantry-sag, are a limitation to the degree of accuracy possible. A possible solution would be to make a dynamical adjustment of the isocenter dependent of the gantry rotation. V. CONCLUSIONS The fact that the isocenter of the ET system has to calibrated using the external lasers introduces systematic deviations between the linac radiation isocenter and the isocenter of the ET system. The deviations were shown to be of an order that may have clinical relevance when considering small clinical margins. Using the EPID and a Winston-Lutz based setup, it is possible to bypass the use of the external lasers and obtain a reduction of the systematic deviations, when using daily imaging. ACKNOWLEDGMENT This work was supported by the Lundbeck Foundation Centre for Interventional Research in Radiation Oncology (CIRRO). The authors have no conict of interest to declare. The authors are solely responsible for the manuscript.
a)

Author to whom correspondence should be addressed. Electronic mail: deta@rn.dk. Telephone: 45 9932 2906; Fax: 45 9932 2904.

Medical Physics, Vol. 39, No. 3, March 2012