|Home | About | Journals | Submit | Contact Us | Français|
In order to accurately assess positioning errors in spinal SBRT, many institutions employ bony-fusion based imaging techniques, such as the ExacTrac™ (Brain Lab) system, in conjunction with 3D verification (performed via CT-on-rails in our practice). We hypothesized that the use of implanted gold fiducial markers could improve the accuracy of patient positioning over bony fusion alone. We addressed this question prospectively, enrolling patients on an IRB-approved protocol. Gold seeds were implanted in the vertebral pedicles flanking the target level. At treatment, setup error was calculated using two methods–standard kV image fusion, and geometric fiducial-based projection, with independent CT-on-rails verification. Analyses of residual set-up error showed that fiducial-based setup agreed with fusion-based determination, but did not significantly reduce error. Offline 6D fusion of the treatment and planning CT illustrated residual rotational error using standard or fiducial based setup. We conclude that the ExacTrac and CT-on-rails platform yields highly accurate results for spinal SBRT setup, with reduced residual error than previously reported. While the addition of fiducials did not further reduce error, the bony fusion approach is now prospectively validated in comparison to implanted fiducials. Both bony fusion and fiducial marker methods are associated with residual rotational error, thus 3D verification remains an important component of spinal SBRT treatment.
Spinal SBRT is a non-invasive treatment for spinal metastases which involves conformal administration of hypofractionated radiation with high precision. A growing body of literature demonstrates the success of this approach, with 1-year local control1-3 reported at ~80-90%, and successful palliation seen in 85-90% of patients.1,4,5 Stereotactic treatment approaches help physically circumvent the limit of spinal cord radiotolerance, thus allowing treatment to higher biologically equivalent doses for radioresistant histologies (e.g., renal cell carcinoma or melanoma6-8), providing more durable local control in the setting of oligometastases, or allowing re- irradiation of well-selected patients who have failed in a prior treatment field.
In order to safely and effectively deliver these treatments, technical sophistication and meticulous attention to detail are required.9-12 Several groups have published the details of their approaches, which use two-dimensional alignment techniques, followed by three-dimensional imaging and daily verification with close support from clinical, therapy and physics staff. In order to reduce the time and effort necessary for treatment, advanced patient positioning systems, such as the ExacTrac™, have been incorporated into spinal SBRT platforms. This tool, currently in use at our institution, uses bony fusion of oblique kV images to assess and correct setup error at the time of treatment.
Previous work13 has sought to characterize the accuracy of the ExacTrac system for this application. Offline 6D-6D fusion (comparison of two CT datasets with the ability to match the images along all 6 axes of movement) of cone-beam CT (CBCT) datasets was used to retrospectively characterize residual setup error. This study found that, while the ExacTrac system results agreed well with the CBCT at the level of 3D-3D fusion, comparison with a 6D fusion technique revealed up to 3.5 mm residual translational error and up to nearly 3° residual rotational error, and advised caution in the use of ExacTrac for SBRT set-up guidance in the absence of 3D verification.
While important, the process of 3D-3D verification at the time of treatment is time-consuming, impacting both staffing needs and stability of patient set-up secondary to discomfort. Drawing on prior experiences14-17, we hypothesized that the integration of implanted fiducial markers into our spinal SBRT workflow might prove sufficiently accurate as to obviate the need for 3D-3D match confirmation. The ExacTrac system provides an alternative marker-based setup method in which marker position can be established based on back projection geometry of the fiducials. As this is a deterministic process, noise and random error in the orthogonal images have less of impact on offset calculation. Thus, we hypothesized that a fiducial based setup might be more accurate for spinal SBRT. Seeking to address this question, and characterize the accuracy of our current platform, here we report the results of a prospective clinical protocol assessing the utility of implanted gold fiducial markers in spinal SBRT.
Following multidisciplinary evaluation, patients dispositioned for spinal SBRT to a single level between T4 and L4 were evaluated for inclusion on a prospective IRB-approved study (NCT01624220). The protocol excluded patients with spinal hardware or prior kyphoplasty at the treatment VB, plus or minus one level, as the surgical hardware or cement artifact might impact the fusion results. Patients with > 50% height loss of the involved VB were also excluded, as were patients who were unsuitable candidates for seed placement (requiring general anesthesia or having uncontrolled bleeding diathesis, INR > 1.7 or platelets < 75 k/µl). An MRI of the spine was performed within 4 weeks of registration, as part of typical treatment planning for spinal SBRT.
Prior to CT simulation, patients underwent fiducial implantation as an outpatient interventional radiology procedure. A co-axial technique, with a 13-gauge vertebroplasty needle (Cook Inc., Bloomington, IN), was used for all procedures. Conscious sedation was administered to all patients. With the patient in the prone position, limited CT images of the spinal level to be irradiated were acquired. Under intermittent CT-guidance, the needle was introduced into the bilateral pedicles of the VB above and the VB below the level to be treated. One 5 mm gold seed (Best Medical International, part #351) was inserted into each needle and placed into the pedicle, resulting in a total of 4 markers (2 above and 2 below). CT verification was performed to verify implantation post-procedure (Figure 1). Seed stability was verified for the first two patients treated; measurements of 8 implanted markers (4 per patient) showed an average measured deviation of 0.5 mm, monitored over 2 weeks following implantation.
Patients with prior irradiation at the level received fractionated SBRT to 27 Gy, delivering 9 Gy x 3 fractions to the gross tumor volume (GTV) and 21-24 Gy to the clinical target volume (CTV). Radiation-naïve patients received 16 Gy to the CTV, and either 18 Gy (for radioresponsive) or 24 Gy (for radioresistant histologies, e.g. renal cell carcinoma) to the GTV, delivered using a simultaneous integrated boost in a single fraction, with radiosurgical target definition as previously described.18,19 For simulation and treatment, a stereotactic BodyFix cradle was employed, as previously described.10 The treatment planning CT was acquired using a Philips Acqsim CT scanner using a 50 cm field of view and 1 mm slice thickness. Treatment was planned using IMRT class solutions20 in Pinnacle3 (version 9.0) [Philips Healthcare, Andover, MA], and delivered using a Varian Clinac 2100 linac with a 5-mm width Millennium MLC.
As part of standard workflow, final patient positioning is verified using a CT-on-rails system (GE Smart GantryTM) to perform an independent 3D verification of patient position. The CT system is coupled to the treatment machine at the opposite end of the vault from the linac (Figure 2). The couch is rotated 180 degrees for CT acquisition, and images are acquired my moving the CT gantry on the rail longitudinally while the patient table is stationary. The linac isocenter is marked with BBs using the in-room laser coordination system(verified by physics staff to be within 0.3 mm of the laser crosshairs), allowing registration of the coordinate systems. The CT-on-rails system performs an independent 3D verification of patient position.
Patient setup for treatment delivery was accomplished using the Brain Lab ExacTrac targeting system.13,21 At the time of treatment, patients were setup in their immobilization device using the marked isocenter coordinates. Initially, setup verification was performed using our standard procedure. Floor-mounted kV imagers, calibrated radiographically to align with the radiation treatment isocenter using the Winston-Lutz method commonly applied in linac-based cranial radiosurgery22, acquired a pair of stereotactic oblique films, which were used by the system software (version 5.5) to compare the patient’s position relative to the treatment planning dataset on the basis of bony anatomy. The kV images are compared to a catalog of digitally reconstructed radiographs (DRRs) derived from the CT simulation dataset, and a software algorithm relies on image fusion of bony structures to generate the necessary couch correction. In-house software is then used to perform a 3D-3D fusion of the reference and treatment CT datasets, which allows computation of the deviation from the planning CT, with tolerances for translation and rotational error of 1 mm and 2 degrees, respectively. The manufacturer does offer a robotic couch system which can execute corrections in 6 dimensions (3 translational – x,y and z, 3 rotational – roll, pitch and yaw), however this robotic couch system was not FDA 510K cleared for the CT-on-rails-system at our institution. As a result, only 3D translational shifts were corrected and verified for treatments. The rotational error, as measured by ExacTrac, would be corrected manually by patient repositioning, should the rotational error tolerance be exceeded.
For patients treated on protocol, an additional step was performed prior to the acquisition of the CT-on-rails verification. Following the standard calculation of isocenter shifts using the bony fusion method, a separate module within the ExacTrac software was used to identify the fiducial markers on the kV images. The center of the seed can be entered using the auto-detection tool provided in the software suite, or manually digitized on the image display screen via mouse input. These coordinates allowed the calculation of an independent estimate of residual translational and rotational error, not dependent on the fusion information. Residual error from both fusion and fiducial based methods were recorded. Root-mean-squared (RMS) analyses and paired-t-tests were used as appropriate to assess the differences in residual error between the two methods.
While the procedure described above allows for verification of patient position at the time of treatment, this analysis in near real-time is limited to 3D fusion (using only translational dataset manipulation tools). To extend our analysis, the verification CT and simulation CT datasets were imported into Pinnacle, where a 6-dimensional (translational and rotational) fusion was performed offline, after patient treatment was completed, providing a more accurate determination of residual error. The ability to use 6 degrees of freedom in this analysis gives a better estimate of the residual error, but is too time-consuming for use during treatment. Importantly, the CT-on-rails system does not have an intrinsic isocenter registered with the linac radiation isocenter, although a laser based fiducial transfer method is used by physicist staff to register the coordinate system prior to CT, and the overall error in isocenter transfer is estimated to be within 1, 0.5 and 0.5 mm for x, y and z, respectively. For this reason, the difference in “true” vs. residual translational error cannot be accurately assessed with this method however, the rotational errors can be assessed accurately. At the time of treatment, the ExacTrac residual error measurements were recorded and subsequently subtracted from the final 6D-6D error measurements as determined by offline analysis in Pinnacle. RMS analyses and paired-t-test were again used to assess whether the two methods might differ in the amount of residual error.
Seven patients were enrolled on the study – 5 patients had thoracic spine lesions and 2 patients had lumbar spine lesions. 5 patients received single fraction spinal SBRT, and two received 3 fraction treatment, totaling 11 patient setups for analysis, summarized in Table 1. Fiducial placement was as per protocol for 6 of 7 patients; in one patient, ambiguous vertebral anatomy led to placement of fiducials one level offset from the specified plan. Spinal fiducial placement was well tolerated, with no adverse events.
As described above, at the time of treatment, deviation of the treatment isocenter from the planning isocenter was independently calculated from kV acquired images using the fusion and fiducial methods. Overall, these two measurement techniques were in very close agreement – the absolute average difference in calculated set-up error between the two methods was 0.24 mm for translational measurements, and 0.32° for rotational measurements. Plots of the discrepancies between the two methods reveal that the distributions are centered about zero (Figure 3), indicating a lack of systematic discrepancy, and are very tightly grouped, with 97% of the observations falling within +/- 1 mm or 1° for translational and rotational error, respectively. RMS analysis confirmed close agreement in the measurements, with no statistically significant difference found for either type of error (ptranslational=0.37, protational=0.26)
As described above, offline comparison of the CT-on-rails dataset to the treatment CT scan was used to provide the best possible estimate of actual rotational error at the time of treatment. Calculation of the discrepancy in rotational error between the 6D-6D fusion and the ExacTrac bony fusion method revealed that there was indeed rotational error that is not captured by the standard ExacTrac fusion algorithm. The average absolute residual rotation error was 0.43°, 0.77°, and 0.69° for roll, pitch, and yaw, respectively. As illustrated in Figure 4, while there is no evidence of systematic error, the discrepancy between the ExacTrac fusion calculation and the “true” rotational error was noted to exceed 1° in ~20% of measurements. As might be expected from the close agreement in residual error measurements described above, RMS analysis shows that the use of fiducial markers does not yield a significantly improved result vs. the fusion method (RMSfusion = 1.13, RMSfiducials=1.04, p=0.36)
In recent years, spinal SBRT has seen increasing adoption, with ~40% of respondents in a practice pattern survey reporting its use in 2010, vs. ~12% in 2005.23 While spinal SBRT is supported by maturing follow-up data, its widespread clinical adoption is still relatively recent, and, as such, there are still a multitude of questions as to best practices for use of this modality. One limitation of spinal SBRT is the technical expertise required to safely and effectively administer the treatment. Treatment delivery systems employ stereotactic immobilization with image guidance, and require meticulous attention to detail, as an optimal patient setup will be accurate within 1 mm of translational error and within 2 degrees of rotational error.24
The increasing adoption of spinal SBRT across a variety of practice settings, while supported by growing clinical literature, underscores the need for rigorous understanding of patient set-up uncertainty associated with stereotactic treatment platforms. To this end, our study aimed to use implanted gold fiducials to both characterize the accuracy of our current ExacTrac-based platform and understand the extent to which fiducials might improve set-up.
While implanted fiducial markers have previously been used for spinal SBRT (e.g. CyberKnife based treatment4), there has not been extensive study comparing their utility to other image guidance techniques. In 2008, a pilot study investigated the use of fiducials in three patients treated with the Novalis Body system, suggesting that the use of implanted fiducials provided superior set-up as compared to bony fusion with the ExacTrac software.14 A limitation to this study was a lack of 3D verification, such that the reported differences between the two systems ultimately represent variation between two algorithms, and are not indexed to an independent assessment of the “true” error. In the current study, the CT-on-rails addresses this limitation and is performed using updated versions of the imaging and software platforms; our results demonstrate very close agreement between the fusion and fiducial methods.
The question of residual rotational error (i.e. rotational error detected by offline 6D-fusion that was not evident during real-time evaluation) has been addressed by a prior study which performed similar 6D-6D offline fusion of CBCT datasets, in addition to phantom studies. In this study, the range of unmeasured rotational error was described to range from +2° to nearly 3°. A comparison to our data suggests a reduced magnitude of variation in the present study, with ~80% of error calculation falling within +/- 1°, and a range of +1.75° to 1.38°, although direct comparison is difficult.
With the analysis presented herein, we have characterized the extent of residual rotational error using a particular technology package (ExacTrac bony fusion, CT-on-rails, standard treatment couch that can execute 3-dimensional corrections). Of note, we have recently upgraded the treatment platform for our spinal SBRT platform to include an updated ExacTrac fusion algorithm (version 6.0), CBCT, and the 6D robotic couch. While our results show that implanted gold fiducials were unable to improve upon the error detection using bony fusion, and do not eliminate uncorrected rotational error, future work will characterize the capabilities of this new platform, including an expansion of our analysis to include tumors of the cervicothoracic junction.
At present, the results of our study bolster confidence in the patient positioning information provided by the ExacTrac bony fusion algorithm, as this non-invasive imaging technology provides results that are indistinguishable from those requiring the (minimally) invasive seed implantation procedure. At the same time, these results again demonstrate that unmeasured rotational error exists in patient set-up, using either method, and underscore the need for 3D verification in the delivery of spinal SBRT.
Authors’ disclosure of potential conflicts of interest
The authors reported no conflict of interest.
Author contributionsConception and design: David C. Weksberg, James N. Yang, Eric L. Chang, Amol J. Ghia, Paul D. Brown.
Data collection: David C. Weksberg, James N. Yang, Xin A. Wang, Zhongxiang Zhao.
Data analysis and interpretation: David C. Weksberg, James N. Yang, Xin A. Wang.
Manuscript writing: David C. Weksberg, James N. Yang.
Final approval of manuscript: all authors.