A retrospective analysis of setup and intrafraction positional variation in stereotactic radiotherapy treatments

Abstract Purpose The aim of this study was to provide a comprehensive assessment of patient intrafraction motion in linac‐based frameless stereotactic radiosurgery (SRS) and radiotherapy (SRT). Methods A retrospective review was performed on 101 intracranial SRS/SRT patients immobilized with the Klarity stereotactic thermoplastic mask (compatible with the Brainlab frameless stereotactic system) and aligned on a 6 Degree of Freedom (DoF) couch with the Brainlab ExacTrac image guidance system. Both pretreatment and intrafraction correction data are provided as observed by the ExacTrac system. The effects of couch angle and treatment duration on positioning outcomes are also explored. Results Initial setup data for patients is shown to vary by up to ±4.18 mm, ±2.97°, but when corrected with a single x‐ray image set with ExacTrac, patient positions are corrected to within ±2.11 mm, ±2.27°. Intrafraction patient motion is shown to be uniformly random and independent of both time and couch angle. Patient motion was also limited to within approximately 3 mm, 3° by the thermoplastic mask. Conclusions Our results indicate that since patient intrafraction motion is unrelated to couch rotation and treatment duration, intrafraction patient monitoring in 6 DoF is required to minimize intracranial SRS/SRT margins.

Approaches for image guidance for SRS can vary between institutions. Typically, alignment to skull bone structures is performed due to the assumed rigid relationship between brain targets and bone anatomy. 3 In linac-based SRS, cone beam CT is often used for visualization of intracranial structures, however, it has limitations at nonzero couch angles. Planar orthogonal imaging at the linac (kV-kV or kV-MV) is also commonly used as it can be acquired from all couch angles, such as with dedicated SRS IGRT systems. One such dedicated IGRT system is the Brainlab ExacTrac Frameless SRS system (Brainlab AG, Germany), which utilizes orthogonal floor-mounted kV tubes and ceiling mounted x-ray detectors.
Over the past decade, several studies have been published investigating patient position stability in stereotactic radiotherapy using stereoscopic, linac-mounted planar or volumetric imaging. Several studies have investigated inter-and intra-fraction patient movement through analysis of pre-and/or post-treatment imaging. 2,[4][5][6][7][8] Further, other studies have provided more granular intrafraction motion data through collection of x-ray images at a range of frequencies during treatment. [9][10][11][12][13] Many however are limited to small patient cohorts or provide only 3 Degrees of Freedom (DoF) or 4 DoF information, or other limited subsets unique to the aims of their experiment. Thus, there are limited data describing the complete picture of patient motion during treatment.
Several opposing findings exist amongst these studies. For example, Badakhshi et al. (2013) found that increased intrafraction imaging frequency is necessary for small margins on account of patient movement 12 and Tarnavski et al. (2016) further added that patient positioning errors become larger with respect to treatment time. 14 Conversely, Lewis et al. (2018) concluded that intrafraction imaging could be reduced after observing minimal patient motion during treatment. 15 In addition, one study found that the type of immobilization masks used in intracranial SRS have a large effect on positional outcomes, 16 while another found that in some cases, no differences were observed. 7 In this study, we perform an offline review on a complete dataset that covers both initial setup and intrafraction motion over a wide range of couch angles in 6 DoF of intracranial SRS and stereotactic radiotherapy (SRT) patients. Our dataset and analysis will enable well-informed further studies in treatment planning and delivery techniques and potential margins. We hypothesize that, due to patient motion, intrafraction image guidance frequency cannot be reduced in order to keep intracranial SRT/SRS margins small.

2.A | Patient Cohort
For this retrospective review we extracted patient setup data from the Brainlab ExacTrac system. Our inclusion criteria was any patient receiving intracranial SRS/SRT on a Varian TrueBeam STx, equipped with the Brainlab ExacTrac image guidance system (v6.2) and Brainlab 6D couch-top, immobilized using the Klarity thermoplastic masks (Klarity Medical Products, OH, USA) in conjunction with the Brainlab frameless stereotactic fixation system. These patients included those treated for intact metastases or surgical cavities. Due to data accessibility difficulties, only data from the last year were included in this study.
Between June 2018 to May 2019, 319 patients received SRS/ SRT at our institution. Of these, 103 (32%) were identified as suitable for this study based on treatment sites and fractionation regimes. Two patients were further excluded from this study due to the use of different immobilization systems, leaving a total of 101 patients for the analysis. The intact metastases cohort contained 87 patients, with 39 patients treated in a single fraction, and 48 in 2-5 fractions (3.4 ± 1.0, µ ± σ). Fifty-eight patients presented a single metastasis, while the remaining 29 had between 2 and 5 metastases (2.6 ± 0.8). The cavity cohort contained 16 patients, with one patient treated in a single fraction, and 15 in 2-5 fractions (3.6 ± 0.9). Fourteen patients each had a single cavity, whilst the remaining two patients had two and three cavities. For this analysis, the intact metastases and postsurgical cavity patient cohorts were combined, giving a total of 258 treatment fractions, in which each fraction represents a treatment with a set of patient positioning data recorded by ExacTrac.

2.B | Treatment planning and workflow
Treatments were planned in Brainlab iPlan (v4.5) (Brainlab AG, Germany) using 1 mm CT slice thicknesses. One isocenter per target volume was prescribed, with 1 mm CTV-PTV margin expansions for intact metastases patients and 2 mm for cavity patients. A dynamic conformal arc (DCAT) technique was used for the majority of cases.
Cases with more complex geometry and within close proximity of organs at risk were planned with intensity modulated radiation therapy (IMRT). For the treatments in this patient cohort, our institution defines action levels for patient positioning with ExacTrac as 0.7 mm, 0.7°. When the patient position is below these values in all 6 DoF, the patient is considered to be aligned; if any translation or rotation is above these values then a correctional shift is applied.
The general image guidance workflow employed during treatment is described in three steps (grouped by treatment workflow).
(1a) Initial patient setup (with non-ionizing radiation). First, with the couch at 0°(C0) and the gantry at 0°, the patients were positioned on the couch and immobilized in the stereotactic mask. The radiation therapist then aligned the patients to the visible in-room lasers, coarsely aligning the target volume to the linac isocenter to within several millimeters. The ExacTrac infrared (IR) system was then used in conjunction with reflective IR markers on the stereotactic mask frame to refine the frame alignment, in 3 DOF, to within 0.7 mm. The alignment of the treatment isocenter to the linac isocenter is now expected to be within 2 mm.
(1b) Initial alignment (with x-rays). With the couch and gantry still at 0°, the first x-ray image set was taken with ExacTrac, the outcome of which describes the accuracy of the initial patient alignment in step (1a). ExacTrac then provided a 6 DoF shift to align the patient to the planned position; this shift, if above the action level, was applied without verification.
(2) First planned couch angle. The couch was rotated to the first planned couch position. The patient was then repeatedly imaged and repositioned until it was verified that all 6 DoF were within the action level. Once the patient position was verified, the first arc was delivered.
(3) Intrafraction image guidance. After delivery of the first arc, the patient received image guided alignment at each planned couch angle. For each alignment, ExacTrac was used to obtain and apply 6 DoF shifts until all 6 DoF were within tolerance.
The generalized treatment workflow was as follows: first, treat with the couch at 0°(C0), and step through from C90 to C270 (anticlockwise) as planned. This workflow, however, was not always strictly followed. In some cases, the first planned couch was nonzero due to either (a) the absence of a planned arc at C0 or (b) iPlan opted for delivering the C0 arc after delivering other arcs first.
During initial patient setup, or during treatment, if the treating staff determined that the verification shifts calculated by ExacTrac were too large (in excess of 5 mm or 2°), or that the patient position was not a suitable match to the planned position, the patient setup was repeated. The mask was removed and refitted, and the setup procedure restarted. If this procedure occurred mid-treatment, the patient was then immediately returned to the previous couch angle at which their alignment was invalidated.
Given the treatment workflow, our analysis of patient motion is divided into two accompanying components. Firstly, we review patient setup and verification shifts (covered in 1a, 1b and 2) and secondly, we review patient intrafraction motion (covered in 3).
Finally, we further investigate the effects of treatment times on patient positioning.  17 Normality tests were performed on the data using the D'Agostino-Pearson K 2 test. 18 Where required, data were reshaped using a sigmoid function to enable the use of statistical tests that require normality. 19 All statistical tests were two-tailed and employed an alpha value of 0.05.

2.C | Data analysis
The mean is defined as µ and the standard deviation, σ. Where box-plots are provided, whiskers illustrate 1.5 times the interquartile range and the black crosses represent outliers. Where 3D information is presented (i.e. a combination of translations or a combination of rotations), the distances are calculated as the modulus of the vector in the three Cartesian axes. Our analysis of confidence intervals was designed to encompass 95% of the sample population, this is calculated as µ ± 2σ. Where asymmetric data are presented, the 95% Confidence Interval (CI) was calculated using the 2.5% to 97.5% percentiles.

3.A | Patient setup
The initial setup of the patient uses no x-ray imaging for alignment and is used to grossly align the patient. The first image set acquired with ExacTrac identifies the overall accuracy achieved during this setup process, the outcomes of which are shown in Fig. 1. These data include all initial setup shift data after patient resets have occurred, as they are also considered a new positioning procedure for the patient.
Normality tests showed that all of the axes were non-normal with the exception of the longitudinal rotation axis (P < 0.05). All translations and the lateral rotation contain either negative or positive asymmetry, which implies a bias exists towards either positive or negative shifts during patient setup. The largest bias is observed in the vertical translations whereby the mean value is −1.03 mm.
A Spearman's Rank Correlation test was used to test for correlations between the six axes (P < 0.05). Two combinations were weakly correlated (0.30 ≤ r s < 0.50). The first of the two observed that as lat-  Normal tests concluded that each of the six distributions in Fig. 2 are shown to be non-normal, extremely narrow and highly skewed.
The distributions are, however, centered within ±0.10 mm and ±0.08°. Again, a Spearman's Rank Correlation test was used to test for correlations between the six axes (P < 0.05). Five correlations were observed in the data, however all of them were considered to be negligible (r s < 0.30), demonstrating inter-axis independency.
Thus, every shift is unique, and a shift observed in one axis does not come with the expectation of a companion shift in another axis.
Again, given the asymmetry present in the data, we have chosen to present the percentile ranges rather than standard deviations. The

3.B | Patient intrafraction shifts
The patient intrafraction shifts, as reported by ExacTrac, are shown in Fig. 3. Normality tests on each axis conclude that each axis is non-normal, presenting with narrow distributions and minimal asymmetry. The mean of each axis is within ±0.11 mm and ±0.08°.
A Spearman's Rank Correlation test was used to test for correlations between the six axes (P < 0.05). Seven correlations were

3.D | Patient time on couch
Intrafraction patient shifts were explored with respect to time and are shown in Fig. 6. Patient treatment times ranged from 4 to 66 min (14 ± 11 min, µ ± σ). In this analysis, it was assumed that patients were never removed from the linac couch when multiple isocenters were treated in the same treatment session. As such, the

3.E | Categorical comparisons
The intrafraction data presented in Fig. 3

| DISCUSSION
The initial patient setup, exclusive of x-ray imaging (Fig. 1), was shown to vary by a large amount between the six axes. The off- the desired 0.7 mm, 0.7°action level. The largest observed shift reduced from 39.24 mm and 6.04°in the initial setup (no image guidance) to 7.20 mm and 5.54°in the first arc alignment phase (image guidance with verification). Given that the maximal shifts do not reduce largely between the initial and verification alignments, we can conclude that applying one correctional shift is not always enough to completely correct the patient position, instead multiple alignments are often required.
The reduction in patient position variation after the initial x-ray acquisition and verification shifts reinforces the necessity for the use of x-ray based pretreatment patient setup in SRS/SRT. Further, this also highlights that although the IR system achieves alignment to within 0.7 mm, that is only for the mask frame. The patient position within the mask can only be corrected with imaging that is capable of correcting based on internal patient anatomy. Only after the patient position is verified, can the IR system indicate potential patient shifts throughout treatment, as promoted by Spadea et al. 8 Although, we propose that IR monitoring of the mask has significant limitations in situations where single isocenters are employed for multiple metastases, as the IR system only registers the patient position in 3 DoF and will not account for any rotation discrepancies.
The patient intrafraction motion data, presented in Figs. 3-6, depict stable patient motion throughout treatment. Figure 3 shows that 95 % of patients will move less than 0.93 mm and 0.75°during their treatment for any given axis, while Fig. 4 shows that the largest intrafraction movement for 95 % of patients will be within 1. Of the studies who provided 6 DoF intrafraction motion data, the magnitude and spread of our data is only in agreement with that reported by Lewis et al. 15 Although not directly comparable, our patient intrafraction shift data are also smaller than that reported by Each study in our literature review of both initial and intrafraction patient shifts presented data only by mean values and various multiples of the standard deviation. It is quite common that the data are neither symmetrically distributed nor statistically normal, and as such, it is not always appropriate to present the confidence intervals by using this method. 19 In our study we have provided 95 % CIs calculated with two methods, using the mean and standard deviation and by using percentiles. The initial setup and verification data presented in Figs. 1 and 2 highlight the limitation of using 95% CIs that are calculated by using the standard deviation to represent the data alone. For example, from Table 1, the longitudinal translations presented for the initial verification shifts (Fig. 2) show that using μ ± 2σ to calculate the 95% CI results in a range of (− In situations where mask material is removed in order to facilitate other means of image guidance (such as optical systems), patient positioning results have been shown to be comparable to that of ExacTrac. 20 Schmidhalter et al. also attributed some positioning inaccuracies to patient weight loss and the subsequent increased movement within the mask, although in our study this is not an issue as we are not treating more than 5 fractions. 11 Occasionally, patient swelling can also occur. In such a situation, an extra "spacer" is placed between the two halves of the patient mask to reduce the pressure on the patient's face. However, given that the swelling reduces the amount of space within the mask, we do not expect that it has a detrimental impact on patient positioning. In severe cases of swelling, the mask may be remade, or more rarely, parts of the mask may be cut away to relieve the pressure. However, given that cutting sections of a mask away for optical surface tracking techniques has shown to provide similar alignment results to that of ExacTrac, 20 we do not expect that such a modification (as rare as it is) would grossly impact our results.
Regarding patient motion with respect to treatment duration, | 117 uncorrected patient displacement from isocenter and found significant correlations between patient intrafraction motion and time. 13 Guckenberger et al. (2012) used positioning outcomes after corrected patient displacement and also found that intrafraction positional errors were significantly correlated with time. 7 Tarnavski et al.
(2016), in an ExacTrac-based study, also found that patient movement (greater than 2 mm or 2°) increased with treatment durations of >10 min. 14  A large caveat of any patient intrafraction motion studies is that the results are intimately tied to the stability of the immobilization system and may also be affected by operator skill and experience (or changes in patient anatomy as previously discussed). Ohtakara et al.
(2012) explored differences between various thermoplastic mask systems and found that patient positioning outcomes is expected to change between masks, 16  Further to this, several statistically significant correlations were observed in our data between patient motion in each axis and tumor displacement. Although all trends were deemed negligible, they could provide a good model for the design and recommendations in future mask immobilization systems.
In our study, patients were planned with either DCAT or IMRT and one isocenter per target which led to, in some cases, treatment durations of up to an hour. However, in recent years, mono-isocenter multi-target VMAT is becoming a popular approach for SRS. 21 Mono-isocenter VMAT techniques offer large reductions in treatment times, however, they also come with challenges in patient positioning. Using the real patient data provided in this study, and several existing studies on the effects of patient position in monoisocenter techniques, 22,23 margins or action levels for image guidance could be inferred.
This work provides a statistically sound overview of patient setup and intrafraction motion throughout SRS and SRT treatments.
The raw data provided and review of our intrafraction data allows for sampling it in future studies where real patient data are required in order to study dosimetric effects of other treatment techniques such as single isocenter, multi-metastases treatments, complete more in-depth studies on the data or provide a time point in which future data can be adequately compared.

| CONCLUSION
We have retrospectively reviewed the variation in our patient setup and intrafraction shifts for intracranial SRS treatments. We

Tomas Kron and Nicholas Hardcastle receive funding from Varian
Medical Systems for an unrelated project.