To frame or not to frame? Cone‐beam CT‐based analysis of head immobilization devices specific to linac‐based stereotactic radiosurgery and radiotherapy

Abstract Purpose Noninvasive frameless systems are increasingly being utilized for head immobilization in stereotactic radiosurgery (SRS). Knowing the head positioning reproducibility of frameless systems and their respective ability to limit intrafractional head motion is important in order to safely perform SRS. The purpose of this study was to evaluate and compare the intrafractional head motion of an invasive frame and a series of frameless systems for single fraction SRS and fractionated/hypofractionated stereotactic radiotherapy (FSRT/HF‐SRT). Methods The noninvasive PinPoint system was used on 15 HF‐SRT and 21 SRS patients. Intrafractional motion for these patients was compared to 15 SRS patients immobilized with Cosman‐Roberts‐Wells (CRW) frame, and a FSRT population that respectively included 23, 32, and 15 patients immobilized using Gill‐Thomas‐Cosman (GTC) frame, Uniframe, and Orfit. All HF‐SRT and FSRT patients were treated using intensity‐modulated radiation therapy on a linear accelerator equipped with cone‐beam CT (CBCT) and a robotic couch. SRS patients were treated using gantry‐mounted stereotactic cones. The CBCT image‐guidance protocol included initial setup, pretreatment and post‐treatment verification images. The residual error determined from the post‐treatment CBCT was used as a surrogate for intrafractional head motion during treatment. Results The mean intrafractional motion over all fractions with PinPoint was 0.62 ± 0.33 mm and 0.45 ± 0.33 mm, respectively, for the HF‐SRT and SRS cohort of patients (P‐value = 0.266). For CRW, GTC, Orfit, and Uniframe, the mean intrafractional motions were 0.30 ± 0.21 mm, 0.54 ± 0.76 mm, 0.73 ± 0.49 mm, and 0.76 ± 0.51 mm, respectively. For CRW, PinPoint, GTC, Orfit, and Uniframe, intrafractional motion exceeded 1.5 mm in 0%, 0%, 5%, 6%, and 8% of all fractions treated, respectively. Conclusions The noninvasive PinPoint system and the invasive CRW frame stringently limit cranial intrafractional motion, while the latter provides superior immobilization. Based on the results of this study, our clinical practice for malignant tumors has evolved to apply an invasive CRW frame only for metastases in eloquent locations to minimize normal tissue exposure.

immobilization. Based on the results of this study, our clinical practice for malignant tumors has evolved to apply an invasive CRW frame only for metastases in eloquent locations to minimize normal tissue exposure.
cranial stereotactic radiotherapy, frame-based immobilization, precision frameless stereotaxy, radiosurgery 1 | INTRODUCTION Accurate treatment positioning and patient immobilization is of the upmost importance for cranial stereotactic radiosurgery (SRS). 1 Traditional approach is to use invasive head frame which can fix the cranium rigidly and no treatment margin beyond the treated target is needed. [2][3][4] With noninvasive head immobilization devices, common practice has been to apply a small planning target volume (PTV) margin to account for uncertainties. Depending upon the immobilization device and image-guidance system that is used to detect and correct for motion, PTV margins typically range from 1 to 3 mm. The clinical disadvantage of adding any PTV margin is an increased risk of radionecrosis as a result of a greater volume of normal tissue receiving the high dose. 5 The drawbacks of using an invasive frame include patient anxiety, pain associated with placement of the screws which are typically applied to the outer table of the cranium, and risk of bleeding and infection at the site of placement. 6 Compared to head frames, noninvasive immobilization systems such as thermoplastic masks have been shown to offer patient immobilization inferior to what is required for SRS, but sufficient for fully fractionated stereotactic radiotherapy (FSRT). [7][8][9][10][11][12] These immobilization systems have been increasingly used in hypofractionated stereotactic radiotherapy (HF-SRT), stereotactic radiation delivered in 2 to 5 fractions, but the performance has not been well studied. 13 Head immobilization technology has made significant advances by incorporating sophisticated mouth bite apparatus 14,15 and integrated vacuum suctioning system that (a) reduces air gaps between the hard palate and the mouth bite to reduce potential slippage and (b) warns the therapists if the patient head moves which causes the vacuum pressure to drop. While there exist a number of studies that have assessed the positional accuracy and stability of different invasive and noninvasive immobilization systems for cranial stereotactic radiotherapy, 6,9,[16][17][18][19][20] to our knowledge, no study has directly compared the performance of various systems used in the same institution.
The aim of the present study is to evaluate patient intrafraction head motion for a series of widely available invasive and noninvasive immobilization systems for SRS, HF-SRT, and FSRT. Congers, NY, USA). 12,18,21 Note that CRW was considered the "gold standard" in this study and the PinPoint system (shown in Fig. 1) was specifically acquired as an alternative to using the invasive CRW frame for SRS at our center. This device is equipped with a vacuum fixation bite-block device consisting of an external and internal component that work in tandem such that patients cannot move their head without losing suction. The internal component contains a patient-specific dental mouthpiece with continuous mild vacuum suction to the upper hard palate (assures tight contact). The external component consists of the dental mouthpiece secured to a metal arch frame that is in turn locked into a carbon-fiber couch board equipped with a thermoplastic head support formed by creating an impression of the back of the skull (note that previous versions utilized an alphacradle for head support 14 ). A reference box with three embedded spherical radiopaque markers is attached over the bridge of the Pin-Point system to assist patient setup.
The distance between the CBCT isocenter and the MV isocenter is measured through daily quality assurance tests and is stringently kept to within 1 mm. All SRS patients were treated on the same unit using stereotactic cones (Elekta AB, Stockholm, Sweden) externally mounted on the gantry head. For all treatment plans, coplanar and noncoplanar beam arrangements were used. The total treatment delivery time was on the order of 20 min for both HF-FSRT and FSRT, and 20 min per isocenter for SRS (with each target having between 1 and 3 isocenters).

2.C | Image-guidance protocol
An initial CBCT scan was acquired after the patient was immobilized and positioned at the treatment isocenter using a localizer device (CRW), a setup reference box (PinPoint) or reference marks on the frame (GTC) and thermoplastic mask (Uniframe, Orfit). Registration to the planning CT, to determine the precise vector shift required to match the CBCT isocenter to the planning CT isocenter, was completed using grayscale matching. CBCT/CT image fusion accuracy has been shown to be <0.1 mm in each direction using the grayscale algorithm 21 and a region of interest encompassing the PTV and cranium (Elekta X-ray Volume Imaging (XVI) software v.4.0). Translational (X = lateral, Y = superior-inferior, Z = anterior-posterior) and rotational (pitch, roll, yaw) offsets were recorded and fine positioning corrections prior to treatment were made using the 6-DOF Hexa-POD robotic couch (Elekta AB, Stockholm, Sweden). Consistent with what others have achieved, 21 HexaPOD was tested during commissioning and found to agree with software to within AE0.3 mm for all translations and AE0.2°for each rotational axis.
For this study, the repositioning threshold was strict at 1 mm and 1°in any translational or rotational axis, respectively since it has been shown that positioning the target to as close to the intended position as possible, that is, 1 mm threshold for patient repositioning, reduces subsequent out-of-tolerance motions and improves the overall precision in delivery. 22 A pretreatment verification CBCT was taken to confirm that the isocenter was within a three-dimensional (3D) vector magnitude of 1.5 mm of the pretreatment CT isocenter.
At the end of treatment, the couch was returned to 0°and a posttreatment CBCT acquired and registered with the treatment planning CT. All shifts were documented and a 3D vector positioning error was quantified using Eq. 1.
For patients that were positioned within our 1 mm/1°tolerance based on pretreatment CBCT (and no pretreatment shifts applied), we used the difference between pre and post-treatment shifts as the intrafractional motion. For the small subgroup of patients that were outside our 1 mm/1°tolerance and that required a second couch shift prior to treatment, the shifts generated from the posttreatment CBCT were used as a surrogate of intrafractional motion.
It is acknowledged that some portion of these shifts included the residual error of couch motion.
The 3D intrafraction displacement was calculated as the vector difference between pre and post-treatment CBCTs. Note that time between pre and post-CBCTs was 20 min and this was the same for both HF-FSRT and FSRT (the total treatment time) and SRS (the time to treat a single isocenter). Although we pooled all of the data together, regardless of treatment technique (SRS/FSRT/HF-SRT) and disease indication, the CRW frame was only used to immobilize brain metastases patients receiving SRS. GTC and thermoplastic masks were used to immobilize brain metastases or primary brain tumor patients receiving FSRT or HF-SRT.

2.D |
The PinPoint system was originally evaluated in FSRT patients and subsequently used for single fraction SRS of brain metastases.

2.E | Statistical analysis
For each immobilization device, the translational, rotational and calculated 3D positioning error from each treatment fraction was, respectively, grouped within the following stages of observation: initial setup, pretreatment, post-treatment, and intrafractional motion.
The mean and standard deviation were calculated from the entire set of all fractions tabulated within the respective translational, rotational, and 3D displacement error datasets.
Box plot was used to describe the distribution of positioning setup displacements for varying immobilization systems as well as for the 3D error. Line plot was also used to show the distribution of inverse cumulative frequency of intrafractional motion by device.
Considering each treatment fraction as independent measurement, an analysis of variance (ANOVA) was used to compare on the differ-

3.A | Initial positioning setup errors
Initial positioning setup errors were analyzed to reveal the setup uncertainty of manual positioning of the different immobilization systems. The 6-DOF translational and rotational displacements for each immobilization system are shown in Fig. 2 as standard box plots that display the full range of variation (from min to max), the likely range of variation (the interquartile range between first and third quartiles), and the typical value (median). Across all immobilization systems, the mean translational displacement in the lateral in Fig. 2(a), superiorinferior in Fig. 2(b) and anterior-posterior in Fig. 2(c) directions together with the mean rotational displacement in the pitch in Fig. 2(d), roll in Fig. 2(e) and yaw in Fig. 2(f) directions were found to be statistically significant but nonspecific.
The lowest initial 3D setup error was observed with the CRW frame (mean value of 0.67 mm) followed by the PinPoint system (mean values of 2.06 and 2.15 mm for SRS and HF-SRT). The GTC frame and the frameless thermoplastic Uniframe and Orfit masks had the greatest mean 3D error and departure from zero (approximately double that of CRW and PinPoint) as summarized in Fig. 3(a). A further pair-wise comparison to the gold standard CRW frame indicated that these results were statistically significant (adjusted P-value <0.0025).

3.B | Pretreatment residual errors
Due to the strict repositioning threshold set at 1 mm and 1°in any translational or rotational axis, respectively, all pretreatment 3D errors as shown in Fig. 3(b) were <1.5 mm (our cutoff for acceptability ensuring that anatomy was within this 3D vector distance of the CBCT isocenter).

3.C | Post-treatment residual errors and intrafractional motion
Post-treatment mean translational residual errors were between À0.25 and 0.11 mm, and the mean rotational errors were between À0.20°and 0.33°for all devices. The differences were found to be significant in all 6-DOF except in the lateral direction. Note that commissioning of the 6-DOF robotic couch showed that the ability to reproduce was only 0.3 mm and 0.2°for all translational and rotational axes, respectively; hence, some of this could be folded into residual errors that were quantified using the post-treatment CBCTs.
A further pair-wise comparison to the gold standard CRW frame indicated statistical significance in the pitch direction for PinPoint HF-SRT, GTC frame, Uniframe, and Orfit (adjusted P-value <0.02).
The post-treatment 3D error, as shown in Fig. 3(c), revealed that the PinPoint and CRW frame had a similar and reduced variability compared to GTC frame, Orfit, and Uniframe. Figure 4 summarizes the intrafractional motion data for all translations and rotations within each immobilization system, and Table 1 summarizes the corresponding mean displacement and 1 standard deviation. In the pitch direction in Fig. 4(d), the mean displacements were <0.12°and variability within AE1°. The largest amount of variability in both the roll in Fig. 4(e) and yaw in Fig. 4(f) directions was observed with the Orfit and Uniframe.
In the lateral direction in Fig. 4(a), there was no significant difference in the mean values, which for all devices were found to be near 0 mm except for the CRW frame which had a mean value of 0.11 mm, but showed the least amount of variability (PinPoint SRS exhibited the same). In the superior-inferior direction in Fig. 4(b), PinPoint for both SRS and FSRT were both found to have the largest amount of mean residual error, À0.17 and 0.11 mm, respectively, but this was statistically nonsignificant. In the anterior-posterior direction in Fig. 4(c), for all devices the mean values were near 0 mm (P-value <0.0001).
With respect to mean intrafractional motion observed in the various systems (see Table 1   In this study, we also observed that the mean intrafractional motion was found to be smaller in the SRS cohort of patients  5). Some of the differences may be attributable to HF-SRT patients becoming more relaxed from the second fraction onwards.
It is also possible that as the number of treatment fractions increases, the relative stability of the patient-specific thermoplastic head support and mouthpiece in restricting head motion lessens.
However, immobilization is still clinically acceptable since the amount of intrafractional motion exceeding 1.25 mm was 0% and  The GTC relocatable head frame had a mean intrafractional motion comparable to that of PinPoint. This is not unexpected since both of them are similar in design such that they are both based on a bite-block device. However, with PinPoint's bite-block, a gentle vacuum suction is applied between the dental mouthpiece and the upper hard palate to assure tight contact, and if the seal is broken a loud hissing sound is heard to alert both the patient and radiation therapist. This added mechanism to ensure stability likely explains the significant reduction in translational and rotational variability during initial setup (Fig. 2) and intrafractional motion (Fig. 4). From The largest initial setup error and intrafractional motion was observed with both thermoplastic mask systems, Uniframe and Orfit.  Elekta, Stockholm, Sweden). 15 Similar in design to PinPoint, eXtend is a noninvasive vacuum bite-block repositioning head frame. Utilizing CBCT, the mean intrafractional motion was found to be 0.4 AE 0.3 mm.
In order to ensure the high accuracy and precision required for SRS particularly when frameless-based localization and fixation is used, both our results and those from literature 15,25 indicate that 3D image-guidance is essential.

| CONCLUSIONS
PinPoint and CRW frame deliver stringent immobilization. Although intrafractional cranial motion is also observed with the invasive CRW frame, it yields the least amount of intrafractional motion and provides superior head immobilization compared to PinPoint and maskbased immobilization systems. The latter noninvasive systems always require image-guidance verification in order to ensure the high accuracy and precision needed for SRS. When deciding to frame or not to frame, our results influenced our practice such that unless lesions are in eloquent tissues where minimal exposure to the surrounding neural tissue is critical, the head immobilization system of choice for SRS is the noninvasive PinPoint.

CONFLI CT OF INTEREST
Dr. Arjun Sahgal has received an honorarium for previous education seminars and research funding from Elekta AB. Drs. Lee, Ruschin and Soliman are participants on a research group sponsored by Elekta AB.