1 INTRODUCTION

Stereotactic body radiation therapy (SBRT) is a safe and effective treatment for spinal metastases, which has been utilized over the past few decades.1, 2 The major advantage of SBRT is that a higher dose per fraction (> 5 Gy/fraction) can be delivered in up to five fractions to yield a greater biological equivalent dose (BED) when compared to conventional radiation. Due to the proximity of spinal tumor targets to the radiation-sensitive spinal cord/cauda, the tumor margin is typically small, and a steep dose gradient is necessary. In addition, spinal tumors often have a concave shape around the spinal canal. To deliver the highly conformal dose to the tumors while avoiding the spinal cord/cauda, patient positioning and immobilization become critical to ensure both adequate tumor coverage and spinal cord/cauda sparing.

SBRT requires adequate patient immobilization and positioning. With the introduction of three-dimensional volumetric imaging (e.g., cone beam computed tomography (CBCT)) and customized body vacuum bags (BodyFIX®), patient set-up accuracy can be achieved within 1 mm at the target isocenter during daily treatment.3-7 BodyFIX® are commonly used in SBRT treatment to maximize repositioning accuracy and intra-treatment patient stability by reducing both involuntary and voluntary patient movement. In addition to BodyFIX®, the treatment couch can compensate for a certain degree of positional errors to better reproduce the patient position at CT-Simulation. Mancosu8 et al. analyzed 2945 CBCTs from brain, lung, liver, pancreas, and prostate treatments and found the maximum absolute values for pitch and roll rotations were 0.7°±0.7° and 0.8°±0.8° in brain, with the lowest values for rotation of 0.4°±0.5° and 0.5°±0.6° in pancreas. Therefore, they suggest that the 6 degrees of freedom (DOF) robotic couch should be used on brain tumors, if available.

However, not all clinics are equipped with a 6-DOF couch needed for the CBCT shift. Often, only a translational correction is available in some machines, like Varian Halcyon (Varian Medical Systems, Palo Alto, CA). While the Varian TrueBeam (TB) STx linear accelerator offers the full 6-DOF correction, the Halcyon machine's couch allows only translational but not rotational correction (i.e., only 3-DOF correction). In addition, when proper target coverage and spinal cord/cauda sparing are of concern, the TB is preferable due to smaller leaf width of 2.5 mm (2.5 mm leaf width is an optional purchase). On the other hand, the Halcyon's MLC collimation system has qualities that makes it clinically desirable,9 despite the concerns of inferior plan quality from the 5 mm effective leaf width. Halcyon has a dual-layer Multi Leaf Collimator (MLC) system. It has a high leaf speed capability of 5 cm/s, compared to 2.5 cm/s in TB, resulting in faster intensity-modulated deliveries with less intra-fraction motion and more efficient patient throughput.

As such, both physical and mathematical efforts have been made to compensate for pitch and roll rotation when a 6-DOF robotic couch is not available. To correct for pitch and yaw setup errors, Boswell10 et al. proposed moving the couch at a very low velocity along the anterior-posterior and left-right axes during Tomotherapy treatment. Several studies have looked at mathematical beam manipulations instead of a physical couch correction. Yue11 et al. used three matrix transformations to adjust for shifts of target isocenter by calculating the necessary shifts in gantry position, couch position, and collimator angles of the beams. The transformed beams will then possess the same positions and orientations relative to the target at treatment as in the original plan. This technique was tested on a head phantom. They found that the target shift was fully corrected in treatment and found excellent agreement in target dose coverage between the plan and the treatment. Fu12, 13 et al. used the same matrix transformation method and reported no dosimetric impact from up to 3° in rotational setup errors in prostate IMRT. They reported that the dosimetric impact for head and neck IMRT cases depended on the proximity of the organ at risk (OAR) to the tumor and that the rotational setup error might need to be routinely corrected on a case-by-case basis. However, these techniques are still in the academic stage and are not yet clinically available.

Other studies14-17 attempted to compensate for rotational errors with a different approach. The studies conducted by Kim14 et al., Oh15 et al., Guckenberger16 et al., and Li17 et al. were based on patient setup errors observed on daily CBCT imaging during treatment. Since the setup errors are different between individuals, the beams are ad hoc recalculated on CBCT for dose variations, and comparisons between individuals are limited due to different combinations of rotational correction.  A more systematic comparison of rotational treatment plans between individuals can be achieved by pooling together data with the same degree of rotation. Study by Peng18 et al. was the only study that simulated patient rotation as opposed to using the rotational errors seen on daily CBCT. This group studied 10 patients with brain tumors who received Intensity-Modulated Radiotherapy (IMRT). They rotated the CT images and contours using their in-house MATLAB script and then performed Dose Volume Histogram (DVH) analysis on the recalculated dose distribution. In most cases, 95% CTV coverage was maintained with rotation setup errors up to 3° for intracranial tumors with a 3 mm CTV-to-PTV margin expansion treated with IMRT. However, for large targets with irregular or elliptical shapes, the target coverage decreased significantly with rotational errors of 5° or more. The results indicate that, even in the absence of translational setup errors, setup margins are needed to account for rotational setup errors. This study did not use translation to correct for rotation, which is slightly different than the standard clinical Halcyon treatment.

A literature search did not show any systematic studies on how to correct patient rotational errors with translation only in spine SBRT treatment, and there is no published data on the magnitude of rotational setup errors we can mitigate with translational corrections alone without compromising the target coverage and exceeding OAR dose limits. This will be very helpful for current Halcyon treatments.

We performed a comprehensive retrospective case-control study on the difference in SBRT plan quality between the Halcyon and Truebeam linear accelerators in the treatment of a single thoracic or lumbar vertebral segment metastasis. In addition, we investigated the dosimetric effect of patient rotational setup errors when only translational corrections are available.

2 METHODS AND MATERIALS 2.1 Patients and volumes

We analyzed a total of 20 patients (9 thoracic, 11 lumbar) with a single-level spinal metastasis, located between T7 and L5 near the spinal canal (Figure 1).  All contouring followed RTOG-0631 guidelines, and on doctor's clinical judgement. We intentionally omitted patients with cervical spine metastases, as the cervical spine has a lot of mobility, and these tumors are safer to treat with a 6-DOF couch. These 20 patients were treated on a TB linear accelerator with a 6-DOF perfect pitch couch and 2.5 mm high definition MLC (HD MLC) at our institution between January 2018 and July 2018.

Axial view of 20 patients with PTV (orange) and cord (cyan) contoured

Figure 1 displays the axial view at the target center and spinal cord or cauda equina for each patient. A 2 mm margin was added from gross tumor volume (GTV) to planning tumor volume (PTV). To account for the potential variations in patient position, the inherent imprecision of image fusion, and the physiological spinal cord/cauda motion and contouring uncertainty due to pixel size, the cord/cauda was either contoured with a 2 mm margin around it or as the entire spinal canal19, 20 (except Pt #5, #9, and #12, who have no margins around the cord/cauda).

The median tumor volume was 52.0 cm3 (17.9–138.7 cm3). The median tumor diameter in the superior-inferior direction was 3.3 cm (2–5 cm) and 4.6 cm (1.7–6.8 cm) in left-right direction. The prescription doses were 12–16 Gy in 1 fraction or 18–24 Gy in three fractions. We scaled all prescription doses to 16 Gy in one fraction in order to average the maximum cord/cauda dose from the plans. All patients were planned in Eclipse AAA V15.6 and treated in the TB linear accelerator with the coplanar volumetric-modulated arc therapy (VMAT) technique. Their treatments were then retrospectively planned for evaluation on the Halcyon with VMAT technique using the same set of planning CT images and contours. No patients were treated on the Halcyon.

2.2 Simulation of patient rotation with respect to isocenter

The dosimetric effect of patient rotation was evaluated by simulating multiple rotational combinations in the Eclipse treatment planning system (TPS) using the AAA algorithm. Since the commercial TPS will not perform calculations on rotated CT image sets with skewed pixels, we used Velocity AI (Varian Medical Systems, Palo Alto, CA) to simulate patient rotations. To do this, we truncated the planning CT exactly 10 cm above and below the isocenter, such that the resultant 3D Image “tilt 0” is centered at the isocenter level in the superior/inferior direction. This image set in Velocity was rotated around the pivot point of the image center in 12 different rotational combinations and defined as 1A-3D in Table 1. The definition of pitch, roll and yaw is illustrated in Figure 2.

TABLE 1. Twelve different rotational combinations 1A 1B 1C 1D 2A 2B 2C 2D 3A 3B 3C 3D Pitch x −1° −1° +1° +1° −2° −2° +2° +2° −3° −3° +3° +3° Yaw y −1° +1° −1° +1° −2° +2° −2° +2° −3° +3° −3° +3°

Definition of pitch, roll, and yaw

Rotational error in roll was not considered in this study, as both the original TB plans and the recreated Halcyon plans were coplanar VMAT with three to four full arcs, and a 1–3° patient roll would not be significant compared to several 360° arcs; each one degree of roll will contribute an error of 1/360, and if a plan has 3 arcs, the contribution will be < 0.1%. This was validated by a pilot study.

The twelve rotated images were imported back to Eclipse, followed by a series of 6-DOF chain registration with auto-matching (Figure 3) between “tilt 0” and 1A 1B 1C 1D 2A 2B 2C 2D 3A 3B 3C 3D. Finally, all contours were overlaid with the fused rotated images after double checking the rotation matrix at technical properties with X = 1° for 1° pitch, Y = 1° for 1° yaw, and so forth.

6-DOF and 3-DOF registration flow chart

We did not rotate the contours in Velocity due to volume variation after multiple manipulations followed by image resampling. Instead, we copied the original structures to registered image after a 6-DOF auto matching in Eclipse. We validated the fidelity of the overlaid contours by calculating the original beam on each perfectly registered (6-DOF) CT image set and comparing the DVH to that of the original plan, which were identical.

After confirming the integrity of the contour and rotation, we deleted all the 6-DOF registration, we then performed a new set of 3-DOF chain registrations with translation alone and recorded the translational shift for correcting each rotational combination (Figure 4). This could mimic the clinical situation where no rotational correction, and only translational correction is available, as in the Halcyon treatment scenario. Dose was recalculated with the same MU on the CT image set with both TB and Halcyon plans. This resulted in 24 sets of treatment plans for each patient, with 8 subsets per 1°, 2° and 3° rotated images respectively.

Rotational error projection at distance

2.3 Plan evaluation

All plans are evaluated pairwise (paired 2-sided t-test) using dose-volume histograms (DVHs). PTV coverage was evaluated using the Conformity Index (CI), Gradient Index (GI) and the Homogeneity index (HI) as follows: 

RTOG Conformity Index: CI = PIV/TV; PIV is the prescription isodose volume; TV is the target volume.

Paddick Conformity Index:  pCI = TVPV2 / (TV × PIV); where PIV is the prescription isodose volume, TVPV is the target volume within the prescribed isodose surface, and TV is the target volume.

Paddick Gradient Index: pGI = PhV/PIV; where PhV is the volume of half the prescription isodose; PIV is the volume of the prescription isodose line. For a plan prescribed to the 80% isodose line, it is the ratio of the 40% isodose volume to that of the 80% isodose volume.

RTOG Homogeneity Index: HI = Dmax/Dp; where Dmax is the maximum point dose and Dp is the prescribed dose to the target volume.

To quantitatively measure the PTV coverage and OAR sparing of each plan, an in-house Eclipse script was developed to extract data from the DVH. We compared PTV coverage, CI, pCI, pGI, and HI.

To compare OAR sparing, we collected the V8Gy, V10Gy, and V12Gy, as well as the D0.03cc to represent maximum OAR dose. The total monitor units (MUs) per fraction and MLC segments were compared to estimate the plan complexity for each plan.

2.4 Statistical analysis

Data were tabulated as mean ± standard deviation. Statistical analysis was performed utilizing Microsoft Excel for paired two-sided t-test. For plan quality, the Halcyon plan was compared with TB plan. For delivery accuracy, the translation corrected plan was compared to the perfect-setup plans in both TB and Halcyon. A value of p = 0.05 or below was statistically significant.

3 RESULTS

Rotational error in roll was not considered in this study, we assumed if a plan has three arcs, the contribution will be < 0.1%. This was validated by a pilot study. Results showed the dosimetric influence of patient rotational setup error was statistically insignificant for rolls of 2° (Table 2).

TABLE 2. Comparison of dosimetric indices of roll 2 and P1 R1 Y1.5 on Halcyon Variable Roll 0 Roll 2 P1 R1 Y1.5 RTOG Conformity Index (CI) 1.12 ± 0.33 1.12 ± 0.33 1.12 ± 0.33 Paddick Conformity Index (pCI) 0.78 ± 0.12 0.77 ± 0.12 0.77± 0.12 Paddick Gradient Index (pGI) 4.08 ± 0.84 4.07 ± 0.84 4.08 ± 0.84 RTOG Homogeneity index (HI) 1.33 ± 0.10 1.34 ± 0.10 1.33 ± 0.10 PTV Coverage (%) 90.8± 5.2 90.7 ± 4.6 90.7 ± 4.6 Max cord/cauda (Gy) 11.9±2.6 12.2±2.4 11.8±2.4 Spinal cord/cauda V8 (cc) 2.2±2.2 2.2±2.2 2.2±2.1 Spinal cord/cauda V10 (cc) 1.5±2.2 1.2±1.8 1.1±1.8 Spinal cord/cauda V12 (cc) 1.1±2.0 0.8±1.5 0.7±1.5 Note: All p-values > 0.05 (n = 20, paired t-test, 2 tails). Roll 0 = 0° roll; Roll 2 = 2° roll; P1 R1 Y1.5 = 1° pitch, 1° roll, 1.5° yaw.

Analysis of the 50 cone beam computed tomography (CBCT) scans showed average rotational setup errors of 0.6°, 0.8°, and 1.2° in pitch, roll, and yaw, respectively. So, we analyzed rotational setup errors of 1°, 1°, and 1.5° in pitch, roll, and yaw respectively. Results showed the dosimetric influence was not statistically significant (Table 2).

At the time of registering rotated images, we recorded the translational shift for correcting each rotational combination (Table 3). Our results showed that without rotational correction, the average absolute translations were 0.2, 0.3, and 0.9 mm in X, Y, and Z, respectively for 1° pitch/yaw combinations, 0.4, 0.4, and 1.7 mm for X, Y, and Z, respectively for 2° pitch/yaw combinations, and 0.5, 0.5, and 2.6 mm for X, Y, and Z, respectively for 3° random pitch/yaw combinations.

TABLE 3. Average translational shift for correcting each rotational combination Rotation X (mm) Y (mm) Z (mm) 1° 0.23±0.38 0.28±0.35 0.94±0.60 2° 0.38±0.52 0.38±0.47 1.72±0.95 3° 0.45±0.70 0.50±0.73 2.55±1.35

Table 4 summarizes the plan quality of TB versus Halcyon. We find that with the same number of arcs, Halcyon plans are comparable to TB with similar target coverage and max cord/cauda dose. However, the average target coverage was about 1.2% lower. Although this was not statistically significant (p = 0.24), it is considered clinically unacceptable. After adding one additional arc, we achieved acceptable plans reviewed by physicians with similar target coverage (92.3 ± 3.0% vs. 92.4 ± 3.3%, p = 0.82), similar pCI (0.83 ± 0.1 vs. 0.81 ± 0.1, p = 0.17), and similar HI (1.31 ± 0.07 vs. 1.31 ± 0.08, p = 0.84) compared to TB plans. Halcyon plans have higher Gradient Index (3.96 ±0.8) than TB plans (3.85 ±0.7) without statistical significance (p = 0.21). Maximum dose to the spinal cord/cauda was also comparable (11.1 ± 2.8 Gy vs. 11.4 ± 3.6 Gy, p = 0.39), as were the V8Gy, V10Gy, and V12Gy for the cord/cauda. The total MUs for Halcyon (4998 ± 1688) were comparable to TB (5463 ± 2155).

TABLE 4. Comparison of dosimetric indices of Halcyon versus TB Variable TB 2–3 VMAT HAL 2–3VMAT HAL 3–4VMAT RTOG Conformity Index (CI) 1.00 ± 0.1 1.10 ± 0.4 1.10 ± 0.2 Paddick Conformity Index (pCI) 0.83 ± 0.1 0.77 ± 0.1* 0.81 ± 0.1 Paddick Gradient Index (pGI) 3.85 ± 0.7 4.06 ± 0.9 3.96± 0.8 RTOG Homogeneity index (HI) 1.31 ± 0.1 1.34 ± 0.1 1.31 ± 0.1 Total MU 5463±2155 4775±1591 4998±1688 PTV coverage (%) 92.30 ± 3.0 91.10 ± 5.4 92.40 ± 3.3 Max cord/cauda (Gy) 11.10±2.8 11.20±2.5 11.40±3.6 Spinal cord/cauda V8 (cc) 2.10±2.2 2.30±2.1 2.30±1.8 Spinal cord/cauda V10 (cc) 1.10±1.8 1.20±1.9 1.00±1.6 Spinal cord/cauda V12 (cc) 0.60±1.3 0.80±1.5 0.60±1.4 Note: All p-values > 0.05 (n = 20, paired t-test, 2 tails) except *p≤0.05.

We further quantified dosimetric effect of correcting rotation with translation alone on TB (Table 5) and Halcyon (Table 6). In both TB and Halcyon, the dosimetric influence of patient rotational setup error was statistically insignificant for rotations of up to 1° pitch/yaw if corrected by couch translation (with similar target coverage, CI, max cord/cauda dose and V8Gy, V10Gy, V12Gy for cord/cauda). At combined rotations of 2° pitch/yaw, the dosimetric influence of patient rotational setup error was statistically significant for TB, but not significant for Halcyon: target coverage decreased by 0.6% for Halcyon and 1% for TB. Target coverage did not go down significantly for Halcyon until combined rotations of 3°, with an average 1.7% drop of target coverage. Despite a statistically significant drop in target coverage, the cord/cauda maximum dose and V8Gy-V12Gy were comparable.

TABLE 5. Comparison of dosimetric indices on 1°–2° rotated image on TB Variable Tilt 0 Rotation A B C D Conformity Index (CI) 1.04±0.12 1° 1.04±0.12 1.04±0.12 1.04±0.12 1.04±0.12 2° 1.04±0.12 1.04±0.13 1.04±0.12 1.04±0.12 PTV Coverage (%) 92.3±3.05 1° 92.00±3.02 92.00±3.14 92.20±2.85 92.20±3.00 2° 91.30±3.36* 91.50±3.51* 91.30±2.50* 91.30±2.97* Max cord/cauda (Gy) 10.4±1.80 1° 10.40±1.84 10.50±1.90 10.40±1.75 10.50±1.85 2° 10.50±1.91 10.60±2.02 10.80±1.76 10.70±2.02 Spinal cord/cauda V8 (cc) 1.77±1.76 1° 1.70±1.77 1.71±1.75 1.72±1.77 1.74±1.75 2° 1.70±1.77 1.70±1.74 1.74±1.74 1.70±1.72 Spinal cord/cauda V10 (cc) 0.69±1.43 1° 0.67±1.41 0.67±1.38 0.68±1.40 0.68±1.38 2° 0.68±1.40 0.68±1.38 0.70±1.38 0.69±1.39 Spinal cord/cauda V12 (cc) 0.26±0.79 1° 0.27±0.79 0.26±0.76 0.26±0.77 0.25±0.75 2° 0.28±0.80 0.27±0.78 0.27±0.76 0.27±0.77 Note: n = 20, paired t-test, 2 tails; all p-value > 0.05 except *p≤0.05. A = (−1°–2° pitch/−1°–2° yaw), B = (−1°–2° pitch/+1°–2° yaw), C = (+1°–2° pitch/−1°–2° yaw), D = (+1°–2° pitch/+1–2° yaw). TABLE 6. Comparison of Dosimetric indices on 1°–3° rotated image on Halcyon Variable Tilt 0 Rotation A B C D Conformity Index (CI) 1.09±0.24 1° 1.10±0.24 1.10±0.24 1.10±0.24 1.10±0.24 2° 1.10±0.24 1.10±0.25 1.10±0.24 1.10±0.24 3° 1.09±0.25 1.09±0.25 1.09±0.24 1.09±0.24 PTV Coverage (%) 91.1±5.35 1° 91.00±5.55 91.00±5.57 91.20±5.37 91.10±5.64 2° 90.60±5.88 90.50±5.93 90.70±5.52 90.40±5.82 3° 89.30±5.86* 89.20±6.12* 89.90±5.40* 89.20±5.81* Max Cord/cauda (Gy) 11.2±2.47 1° 11.20±2.48 11.30±2.49 11.20±2.49 11.30±2.49 2° 11.20±2.55 11.30±2.58 11.40±2.48 11.40±2.51 3° 11.40±2.61 11.50±2.65 11.50±2.34 11.50±2.58 Spinal Cord/cauda V8 (cc) 2.34±2.14 1° 2.29±2.12 2.30±2.13 2.31±2.12 2.30±2.13 2° 2.28±2.11 2.28±2.13 2.31±2.10 2.28±2.09 3° 2.27±2.12 2.25±2.12 2.29±2.06 2.28±2.08 Spinal Cord/cauda V10 (cc) 1.19±1.86 1° 1.22±1.86 1.23±1.86 1.22±1.82 1.22±1.82 2° 1.21±1.87 1.23±1.86 1.22±1.83 1.21±1.84 3° 1.20±1.82 1.17±1.80 1.18±1.78 1.17±1.77