Abstract
The accuracy of stereotactic body radiotherapy (SBRT) in the liver is limited by tumor motion. Selection of the most suitable motion mitigation strategy requires good understanding of the geometric and dosimetric consequences. This study compares the geometric and dosimetric accuracy of actually delivered respiratory gated SBRT treatments for 15 patients with liver tumors with three simulated alternative motion adaptation strategies. The simulated alternatives are MLC tracking, baseline shift adaptation by inter-field couch corrections and no intrafraction motion adaptation. The patients received electromagnetic transponder-guided respiratory gated IMRT or conformal treatments in three fractions with a 3–4 mm gating window around the full exhale position. The CTV-PTV margin was 5 mm axially and 7–10 mm cranio-caudally. The CTV and PTV were covered with 95% and 67% of the prescribed mean CTV dose, respectively. The time-resolved target position error during treatments with the four investigated motion adaptation strategies was used to calculate motion margins and the motion-induced reduction in CTV D95 relative to the planned dose (ΔD95). The mean (range) number of couch corrections per treatment session to compensate for tumor drift was 2.8 (0–7) with gating, 1.4 (0–5) with baseline shift adaptation, and zero for the other strategies. The motion margins were 3.5 mm (left–right), 9.4 mm (cranio–caudal) and 3.9 mm (anterior–posterior) without intrafraction motion adaptation, approximately half of that with baseline shift adaptation, and 1–2 mm with MLC tracking and gating. With 7 mm CC margins the mean (range) of ΔD95 for the CTV was 8.1 (0.6–29.4)%-points (no intrafraction motion adaptation), 4.0 (0.4–13.3)%-points (baseline shift adaptation), 1.0 (0.3–2.2)%-points (MLC tracking) and 0.8 (0.1–1.8)%-points (gating). With 10 mm CC margins ΔD95 was instead 4.8 (0.3–14.8)%-points (no intrafraction motion adaptation) and 2.9 (0.2–9.8)%-points (baseline shift adaptation). In conclusion, baseline shift adaptation can mitigate gross dose deficits without the requirement of real-time motion adaptation. MLC tracking and gating, however, more effectively ensure high similarity between planned and delivered doses.
Export citation and abstract BibTeX RIS
1. Introduction
In stereotactic body radiotherapy (SBRT), high radiation doses are delivered to ablate well-defined tumor volumes in the human body. SBRT has proven to be an effective treatment modality for both primary and secondary liver tumors (Aitken and Hawkins 2015, McPartlin and Dawson 2016). However, due to the risk of normal tissue damage successful liver SBRT requires a high geometric treatment accuracy (Dawson et al 2002, Mendez Romero et al 2009, Van den Begin et al 2014). This accuracy is limited by intrafraction tumor motion, which can exceed several centimeters and include both respiratory induced motion and baseline drift (Worm et al 2013, Poulsen et al 2014, Bertholet et al 2016, Nankali et al 2017).
A typical strategy to mitigate the effects of intrafaction motion is respiratory gating, where the beam is paused when the target (or a surrogate for the target position) moves away from the planned position (Ge et al 2013). Three different options are available for internal liver tumor motion monitoring during respiratory gating. The first option is direct monitoring of fiducial markers implanted near the tumor using one (Poulsen et al 2014) or two (Shirato et al 2004, Hanazawa et al 2017) kV x-ray imagers. The fiducial markers are used as tumor position surrogates to overcome the poor visibility of liver tumors in x-ray images. The second option uses the correlation between a continuous external surrogate signal and sparse x-ray monitoring of internal tumor motion (Nankali et al 2016), which is commercially available on specialized accelerators such as the Cyberknife (Schweikard et al 2000) and Vero (Kamino et al 2006), but also possible on conventional linear accelerators (Bertholet et al 2018). The third option is the use of electromagnetic transponders implanted near the tumor and monitored wirelessly by an external antenna system (Calypso, Varian Medical Systems, Palo Alto, CA) (Balter et al 2005). The Calypso system has been used in the prostate for more than a decade (Kupelian et al 2007) and received CE and FDA clearance in 2014 for general use in soft tissue.
A promising future option for motion management in liver SBRT is MLC tracking where the treatment beam is adapted to tumor motion by shifting the MLC aperture during treatment delivery (Keall et al 2001). MLC tracking was first proposed in 2001 (Keall et al 2001) and the first patients were treated in 2013 for prostate cancer (Keall et al 2014). MLC tracking can be performed on a conventional linear accelerator unlike commercially available tumor tracking strategies based on robotic (Schweikard et al 2000) or gimbal (Kamino et al 2006) beam alignment. Calypso-guided MLC tracking has been clinically demonstrated for prostate (Colvill et al 2015) and lung cancer (Booth et al 2016) treatments and could prove very valuable for liver SBRT, where target motion is typically larger. The combined use of jaws and a binary MLC for real-time motion tracking during helical tomotherapy was recently demonstrated (Schnarr et al 2018).
A simpler motion mitigation strategy than real time motion adaptation by gating or tracking could be to correct the couch position between each individual treatment field in case of intrafraction baseline drift. It may prevent dosimetrically costly systematic errors (van Herk 2004) in a simpler way without the requirement of real-time motion adaptation. However, selection of the most suitable motion mitigation strategy requires better understanding of the geometric and dosimetric consequences of the different strategies.
The present study is based on a newly established database of the time-resolved liver tumor motion monitored by the Calypso system throughout 45 liver SBRT treatment fractions for 15 patients (Worm et al 2018). This previous study demonstrated the feasibility of Calypso-guided gated liver SBRT with improved treatment accuracy compared to standard non-gated treatment delivery. In the present study, we use the observed intrafraction motion to investigate and compare the geometric and dosimetric accuracy of four motion mitigation strategies including respiratory gating, MLC tracking, baseline shift adaptation by couch corrections between treatment fields and treatments with no intrafraction motion adaptation.
2. Material and methods
2.1. Patients, treatment planning and delivery
The patient population and treatments have previously been described in detail (Worm et al 2018). Key characteristics are summarized in table 1. After providing informed consent, fifteen patients with primary liver cancer or liver metastases received SBRT in three fractions following a clinical trial protocol on Calypso-guided gated liver SBRT. Real-time motion monitoring during treatment delivery was provided by electromagnetic transponders (Calypso, Varian Medical Systems) implanted percutaneously near the tumor under ultrasound guidance. Patient fixation was based on an in-house made frame supporting a vacuum cushion. Abdominal compression was not used. Treatment planning was performed on a contrast-enhanced exhale CT scan with 2 mm slice thickness. The planning target volume (PTV) was defined by expanding the clinical target volume (CTV) by 5 mm in the anterior–posterior (AP) and left–right (LR) directions and by 7 mm or 10 mm in the cranio–caudal (CC) direction. Seven-field IMRT or 3D conformal plans were designed to deliver the prescribed mean CTV dose (table 1) in three fractions. The PTV dose was non-uniform with 95% and 67% of the prescribed dose covering at least 99% of the CTV and PTV, respectively (Blomgren et al 1995).
Table 1. Patient and treatment characteristics.
| Characteristics | Number of patients |
|---|---|
| Diagnosis (and prescribed mean CTV dose) | |
| Primary liver cancer (48 Gy) | 4 |
| Metastases (45 Gy/56.25 Gy/61.8 Gy) | 1/9/1 |
| Implantation needle gauge | |
| 14-gauge/17-gauge | 10/5 |
| Number of transponders used for localization | |
| 2/3 | 3/12 |
| Planning CT scan | |
| Exhale breath-hold/exhale phase of 4DCT | 13/2 |
| Cranio–caudal PTV margin | |
| 7 mm/10 mm | 12/3 |
| Treatment plan | |
| IMRT/3D conformal | 8/7 |
The treatments were delivered in free breathing with respiratory gating on a TrueBeam accelerator with a Millennium MLC (Varian). Transponder localization at 25 Hz by a Calypso 3.0 system was used to turn on the beam only when the centroid of the transponders deviated from the planned exhale position by less than 3 mm in the LR and AP directions and 4 mm in the CC direction (Poulsen et al 2015). Patient setup was performed under Calypso guidance such that the centroid position was within the gating window at end exhale and had a mean value of approximately zero in all three directions during the open gate time intervals. A pre-treatment CBCT scan acquired after the Calypso-guided setup was used to confirm that the transponders were close to their planned position in the exhale phase. During the treatment fraction, the couch was corrected if a baseline drift of the exhale transponder centroid position resulted in a mean centroid position that visibly deviated from zero in any direction during the open gate time intervals. Typically, a baseline drift exceeding 1–2 mm in any direction would trigger a couch correction, but there was no specific threshold in terms of baseline drift magnitude since this quantity was not directly reported by the system but had to be visually identified on the motion curves during treatment. The mean (range) of the duty cycle during the gated treatments was 62.5% (29.1%–84.9%) (Worm et al 2018).
After the treatments Calypso log files with the transponder centroid motion and accelerator trajectory log files were synchronized in order to obtain the tumor motion, couch corrections, and linac parameters during the gated treatments (Worm et al 2018). The log file synchronization was based on temporal alignment of linac beam holds and target motion outside of the gating window, assuming beam gating latencies of 196 ms and 64 ms for beam-on and beam-off, respectively (James et al 2016).
2.2. Motion adaptation strategies
The tumor motion during treatment delivery was used to investigate the geometric and dosimetric accuracy for the following four motion adaptation strategies for all 15 patients and 45 fractions: (1) the actual respiratory gated treatments, (2) simulated treatments with MLC tracking, (3) simulated treatments with baseline shift adaptation by inter-field couch corrections, and (4) simulated treatments without intrafraction motion adaptation.
In the three simulated adaptation strategies, the internal target motion was first found from the recorded Calypso motion by subtracting the couch corrections performed during the actual gated treatments. Next, CBCT-based patient setup was simulated by offsetting the motion trajectory such that it had a mean position of zero during the 60 s pre-treatment CBCT acquisition. Seven patients had no Calypso motion recorded during the CBCT because couch centering to ensure clearance during the CBCT acquisition moved the transponders out of the Calypso measurement volume. Instead, the 60 s motion recorded just before the couch centering was used to represent the motion during the CBCT.
The MLC tracking treatments were simulated with an in-house developed Matlab software program (Toftegaard et al 2017) that models Calypso-guided MLC tracking by the iTools Tracking suite on a TrueBeam accelerator (Hansen et al 2016). The simulator uses the Dicom treatment plan and motion trajectory to simulate all accelerator motions and MU delivery, which are then stored in output log files similar to the trajectory log files from actual TrueBeam accelerator treatments. The MLC tracking simulator had the same tracking system latency (approximately 146 ms) and used the same linear Kalman filter prediction to compensate for the latency as the real TrueBeam MLC tracking system (Hansen et al 2016) (Toftegaard et al 2017). MLC tracking is less efficient perpendicular to the MLC leaves than parallel to the leaves because leaf pairs have to open and close to adapt to perpendicular motion. Since this was not considered when designing the original treatment plans 105 of the 116 fields had an unfavorable MLC leaf orientation that was (within ±20°) perpendicular to the CC direction, which is the dominant motion direction in the liver (Worm et al 2013). Therefore, new plans were generated with the MLC leaves parallel to the CC directions if the MLC tracking simulations showed that sub-optimal MLC orientation hindered efficient MLC tracking. This replanning was performed for two patients (Patients 3 and 5) and was only used for the MLC tracking strategy. Both modified plans fulfilled the clinical plan requirements and had dose distributions that were nearly identical to the original plans.
In the baseline shift adaptation strategy, the mean target position during each field delivery was calculated. If it deviated more than 2 mm from the planned position in any direction the couch was corrected with the mean deviation in all three directions before delivery of the following field. This strategy could be a practical and cost effective way to adapt to intrafraction baseline shifts without the requirement of online real-time motion monitoring.
All simulated treatments assumed the same starting time of each field as in the actual gated treatments in order to have identical motion in all four strategies. This was done although the field delivery would be faster for the three simulated adaptation strategies since no gating is applied, and the pauses between fields would be shorter for MLC tracking and no intrafraction motion adaptation since inter-field couch corrections are avoided.
In the actual gated treatments, twelve patients were planned with 7 mm PTV margins in the CC direction while three patients were planned with 10 mm CC margins, which is our standard for non-gated treatments. For the current study, new plans with 7 mm or 10 mm margin were made such that all patients had plans with both margin sizes. The treatments without motion adaptation and with baseline shift adaptation were simulated with both margin sizes to investigate the dosimetric impact of the margin. The gating and MLC tracking treatments were only simulated with the 7 mm margin since this margin was large enough to maintain CTV doses close to the planned doses.
2.3. Geometric and dosimetric accuracy
For gating, baseline shift adaptation and no motion correction, the time resolved geometric target position error was given directly in patient coordinates as the target position relative to the isocenter. For MLC tracking, the geometrical error was first determined in beam's eye view (BEV) as the 2D shift of the current MLC aperture from the ideal MLC aperture at each time point. Here, the ideal MLC aperture during MLC tracking was defined as the planned MLC aperture at a given time point translated in BEV to the current target position (Poulsen et al 2012a). The shape of the current MLC aperture at a given time may differ from the shape of the ideal MLC aperture due to ongoing MLC leaf adjustments. Therefore, the 2D shift between the two MLC apertures was estimated by generating binary images of the MLC apertures, blurring these images by convolution with a 2D Gaussian kernel, and determining the shift that maximized the normalized cross correlation between the two images. The resulting geometrical error in BEV was then used to estimate the error in patient coordinates. The CC error in the patient was estimated as the CC BEV error during all coplanar treatment fields. The LR error and AP error were estimated as the lateral BEV error during all coplanar fields with vertical gantry angles (within 20°) and horizontal gantry angles (within 20°), respectively. For MLC tracking, the geometrical error along a given patient coordinate was therefore only estimated for a subset of the treatment fields, which was assumed to be representative for the entire treatment fraction.
For each motion adaptation strategy, the geometrical error in patient coordinates was calculated for all beam-on periods during the entire treatment course. Following the formalism of van Herk (2004), the mean mp and standard deviation σp of the patient specific geometrical error were then used to calculate the group mean error GM for the population as the mean of mp, the systematic error Σ as the standard deviation of mp, and the random error σ as the root mean square of σp. Finally, the population based margin needed to account for target motion was estimated as 2.5Σ + 0.7σ.
For each motion adaptation strategy and treatment fraction, the CTV dose was reconstructed by a motion-including dose reconstruction method that models target motion as multiple isocenter shifts (Poulsen et al 2012b) and uses the clinical treatment planning system (Eclipse 13.7, Anisotropic Analytical Algorithm, Varian) to calculate the delivered target dose. A spatial resolution of 1.5 mm was used, i.e. the target motion was modeled by applying the same isocenter shift to all target positions within a 1.5 × 1.5 × 1.5 mm3 cube. The CTV mean dose (Dmean) and the minimum dose to 95% (D95) and 99% (D99) of the CTV were calculated for each treatment fraction and treatment course, and their deviation from the planned CTV doses was reported.
Statistical significances of geometric and dosimetric differences between all four motion adaptation strategies were calculated using Wilcoxon's signed-rank test.
3. Results
Figure 1 shows the tumor motion for simulated treatments with no intrafraction motion adaptation and with baseline shift adaptation at fraction 3 for Patient 1. Three couch corrections were performed during this fraction to compensate for the baseline drift. The mean (range) number of couch shifts per fraction to compensate for tumor drift was 2.8 (0–7) with gating, 1.4 (0–5) with baseline shift adaptation, and zero for the other strategies. In 12 fractions, there was no couch correction with the baseline shift adaptation strategy and thus no difference between baseline shift adaptation and no motion adaptation.
Figure 1. Tumor motion at fraction 3 for Patient 1 without intrafraction motion adaptation and with baseline shift adaptation in the left–right (LR), cranio–caudal (CC), and anterior–posterior (AP) directions. Grey areas mark the time of the pre-treatment CBCT scan while green areas show the time of beam delivery. The three vertical red lines indicate the time of inter-field couch corrections to adapt to baseline shifts. The Calypso signal was temporarily lost as the couch was rotated for non-coplanar treatment fields approximately 900 s into the trajectory.
Download figure:
Standard image High-resolution imageThe geometric errors for all patients and motion mitigation strategies are presented in figure 2 while the resulting motion margins are summarized in table 2. Baseline shift adaptation largely reduced the systematic errors and approximately halved the motion margins compared to no intrafraction motion adaptation. Random errors were not reduced by baseline shift adaptation. Gating and MLC tracking had very similar geometric accuracy with efficient reduction of both systematic and random errors (figure 2 and table 2). In all three directions, the magnitude of the geometrical error was significantly smaller with baseline shift adaptation compared to no intrafraction motion adaption and with MLC tracking or gating compared to baseline shift adaptation (p < 0.001). The geometrical error was significantly smaller with MLC tracking than with gating in the CC and AP directions (p = 0.001), but not in the LR direction (p = 0.19).
Table 2. Systematic and random geometric errors and motion margins.
| Gating (mm) | MLC tracking (mm) | Baseline shift adaptation (mm) | No motion adaptation (mm) | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LR | CC | AP | LR | CC | AP | LR | CC | AP | LR | CC | AP | |
| GM | −0.02 | 0.25 | −0.18 | 0.00 | 0.01 | −0.01 | −0.05 | 0.53 | −0.30 | −0.25 | 2.17 | −0.91 |
| Σ | 0.37 | 0.33 | 0.22 | 0.20 | 0.21 | 0.09 | 0.29 | 0.47 | 0.33 | 1.10 | 2.41 | 0.94 |
| σ | 0.64 | 1.83 | 1.07 | 0.76 | 1.46 | 0.76 | 1.07 | 4.80 | 2.19 | 1.17 | 4.77 | 2.22 |
| PTV | 1.4 | 2.1 | 1.3 | 1.0 | 1.5 | 0.8 | 1.5 | 4.5 | 2.4 | 3.5 | 9.4 | 3.9 |
GM = group mean, Σ = systematic error, σ = random error, PTV = planning target volume margin (2.5 * Σ + 0.7 * σ, valid for 95% PTV prescription level (van Herk 2004)), LR = left–right, CC = cranio–caudal, AP = anterior–posterior.
Figure 2. Mean (±standard deviation) geometric errors accumulated over all three fractions for each patient for the four motion adaptation strategies in the left–right (LR), cranio–caudal (CC) and anterior-posterior (AP) directions.
Download figure:
Standard image High-resolution imageFigure 3 compares the motion-including reconstructed CTV doses with the planned static dose for Patients 1 and 11, who both had large systematic intrafraction tumor drift in the cranial direction at all treatment fractions (figure 2). Since these patients were originally planned with the larger 10 mm CC PTV margin the gating and MLC tracking dose reconstructions were also performed for this margin (and included in figure 3). Without motion adaptation the cranial drift resulted in dose deterioration in the cranial end of the CTV, in particular for Patient 11 due to a small CTV of 1.6 cm3 (figure 3(b)). The CTV dose was partly restored with baseline shift adaptation and fully restored with gating and MLC tracking.
Figure 3. Planned dose distribution in the coronal plane in the center of the CTV (red contour) and PTV (blue), and reconstructed dose distributions for all motion adaptation strategies accumulated over all three fractions for Patients 1 and 11. Dose levels ⩾95% are shown. The numbers specify the reduction in CTV D95 caused by intrafraction motion.
Download figure:
Standard image High-resolution imageThe mean reduction in CTV D95 is shown for each patient in figure 4 while figure 5 shows the mean CTV dose volume histogram for each motion adaptation strategy. Table 3 summarizes the extracted CTV dose volume parameters. As expected from the geometric results, baseline shift adaptation markedly improved the CTV dose coverage for those outlying patients who had large dosimetric errors without intrafraction motion adaptation. Gating and MLC tracking both resulted in good agreement between planned and delivered doses for all patients with slightly smaller CTV ΔD95 with gating than with MLC tracking (table 3, p = 0.05) despite the smaller geometric errors found for MLC tracking.
Table 3. Mean (and range) reduction in delivered CTV dose relative to the planned CTV dose per fractions and per patient (accumulated over all three fractions).
| Gating | MLC tracking | Baseline shift adaptation | No motion adaptation | |||||
|---|---|---|---|---|---|---|---|---|
| Per fraction | Per patient | Per fraction | Per patient | Per fraction | Per patient | Per fraction | Per patient | |
| Percent-point dose reduction with 7 mm cranio–caudal PTV margin | ||||||||
| ΔDmean | 0.5 (−0.1–2.1) | 0.5 (0–1.5) | 0.6 (−0.3–1.8) | 0.6 (0–1.4) | 2.3 (0–11.6) | 2.3 (0.3–10.2) | 3.6 (0.1–20.2) | 3.6 (0.4–14) |
| ΔD95 | 0.9 (0–3.0) | 0.8 (0.1–1.8) | 1.0 (0–2.7) | 1.0 (0.3–2.2) | 4.2 (0.1–15.8) | 4 (0.4–13.3) | 8.4 (0.3–45) | 8.1 (0.6–29.4) |
| ΔD99 | 1.2 (0–4.2) | 1.1 (0.1–2.4) | 1.4 (0.5–3.7) | 1.4 (0.5–3) | 5.5 (0.2–17.3) | 5.2 (0.5–14.8) | 11 (0.5–47.8) | 10.4 (0.9–35.7) |
| Percent-point dose reduction with 10 mm cranio–caudal PTV margin | ||||||||
| ΔDmean | — | — | — | — | 1.6 (0–8.4) | 1.7 (0.2–7.7) | 2.3 (0–12.1) | 2.3 (0.3–10.0) |
| ΔD95 | — | — | — | — | 3 (0–11.3) | 2.9 (0.2–9.8) | 5 (0.1–24.8) | 4.8 (0.3–14.8) |
| ΔD99 | — | — | — | — | 4.0 (0.1–14.8) | 3.8 (0.3–11.7) | 6.9 (0.1–33.1) | 6.5 (0.3–23) |
Figure 4. Mean reduction in delivered CTV D95 dose relative to the planned dose per patient for all motion adaptation strategies.
Download figure:
Standard image High-resolution image
Figure 5. Mean CTV dose volume histogram (DVH) of all treatment fractions for the four motion adaptation strategies. Shaded areas show the 10–90 percentile DVH range.
Download figure:
Standard image High-resolution imageThe reduction in CTV D95 increased markedly with the mean magnitude of the geometric error (figure 6). The mean magnitude of the CC error exceeded 7 mm in eight fractions without motion adaptation and in one fraction with baseline shift adaptation. With 7 mm CC PTV margins ΔCTV D95 exceeded 12%-point for all these fractions (figure 6, left). With 10 mm CC PTV margins ΔCTV D95 exceeded 12%-point for five of the fractions without motion adaptation (figure 6, right).
Figure 6. Reduction in CTV D95 relative to planned D95 versus the mean magnitude of the geometric error in the cranio–caudal (CC) direction during treatment with CC PTV margins of 7 mm (left) and 10 mm (right).
Download figure:
Standard image High-resolution image
4. Discussion
This comprehensive study with 270 motion-including dose reconstructions compared the geometric and dosimetric accuracy of four intrafraction motion adaptation strategies for liver SBRT. The study was based on a highly detailed database of intrafraction internal liver tumor motion measured with Calypso transponders throughout 45 treatment fractions delivered with respiratory gating. The motion patterns included respiratory motion of up to 3 cm and baseline shifts during treatment of up to 1 cm (Worm et al 2018). While motion details and reconstructed doses of the actual gated and simulated non-gated treatments have been published previously (Worm et al 2018) the current study extends the analysis with systematic comparison of 7 mm and 10 mm margin plans for each patient and with simulated deliveries with MLC tracking and baseline shift adaptation in addition to gating and no intrafraction motion adaptation. This is the first side-by-side comparison of these four motion strategies, which are all implementable on conventional linear accelerators for radiotherapy. Previous clinical work on motion adaptive liver SBRT treatments at conventional linear accelerators focused on gating using an external optical signal placed on the patient's abdomen as surrogate for the internal target position (Li et al 2012, Ge et al 2013).
One of the simplest strategies for intrafraction motion adaptation would be to compensate for tumor baseline shifts by correcting the couch position between each individual treatment field to the mean position of implanted fiducial markers. Previous studies using magnetic resonance imaging (MRI) have shown systematic cranial baseline shifts of the liver (von Siebenthal et al 2007), and Patients 1, 3, and 11 in the present study all had similar cranial intrafraction baseline shifts at each fraction. The baseline shift correction proved to be a cost effective method to prevent the large CTV dose degradation that these patients may have without intrafraction motion adaptation (figure 4). Still, some CTV dose deficit remained due to non-compensated respiratory motion and residual drift motion, in particular with the 7 mm CC PTV margin (figures 4–6 and table 3). Gating and MLC tracking both ensured good CTV dose coverage for all fractions even with 7 mm CC margins. Although the geometric errors were smaller with MLC tracking than with gating (table 2) the dosimetric errors for the CTV were slightly larger (table 3). The reason for this could be that MLC tracking in addition to the shift between current and ideal MLC aperture in table 2 also distorts the planned MLC aperture shape when the MLC leaves are adjusted to new MLC aperture positions. This effect is not present in gating. While Calypso-guided gating is FDA and CE approved, MLC tracking is only available in research mode. However, MLC tracking has the advantages of faster delivery with 100% duty cycle, a more automated workflow with no need for couch corrections, and the potential to correct for target rotations and deformation (Ge et al 2014). The ability of both MLC tracking and gating to deliver doses that are robust to intrafraction motion has previously been demonstrated in experiments for lung tumor motion (Falk et al 2014) and in simulations for prostate radiotherapy (Colvill et al 2014).
For the actual gated treatments, the mean time from CBCT start to the end of the last treatment field was 25.2 min (Worm et al 2018). From our clinical experiences with the gating we coarsely estimate that each couch correction adds approximately 1 min to the treatment time. With an average of 2.8 couch corrections per fraction, a mean duty cycle of 62.5% (Worm et al 2018), and mean total beam-on times of 4.8 min (conformal plans) or 8.1 min (IMRT plans) we estimate that gating on average prolonged the treatment time with approximately 5.7 min for conformal plans and 7.7 min for IMRT plans. For comparison, baseline shift adaptation would prolong the mean treatment time with approximately 1.4 min (1.4 couch corrections) while MLC tracking and no intrafraction motion adaptation would add no time to the treatment.
Although real-time monitoring is not needed for baseline shift adaptation, one still has to determine the mean tumor position during each field delivery. A potential solution could be kilovoltage intrafraction monitoring (Poulsen et al 2014), where the 3D position of implanted fiducial markers is determined by the gantry-mounted x-ray imager of a conventional linear accelerator. A relatively low localization frequency may be used since only the mean target position is needed. As an example, a sampling frequency of 0.33 Hz, which is possible by triggered kV imaging on Varian TrueBeam accelerators, would estimate the intra-field mean positions of this study with root-mean-square errors of 0.19 mm (LR), 0.80 mm (CC), and 0.55 mm (AP) and result in PTV margins with baseline shift adaptation of 1.6 mm (LR), 4.7 mm (CC) and 2.5 mm (AP). These margins are only slightly larger than the margins in table 2 for continuous localization (1.5 mm (LR), 4.5 mm (CC), 2.4 mm (AP)).
The baseline shift adaptation in this study used a 2 mm threshold for triggering of couch corrections. This threshold was chosen as a compromise to correct dosimetrically expensive systematic errors caused by baseline shifts with a reasonable number of couch corrections: with an average (range) of 1.4 (0–5) couch corrections per fraction the margins were reduced from 3.5 mm (LR), 9.4 mm (CC) and 3.9 mm (AP) to 1.5 mm (LR), 4.5 mm (CC) and 2.4 mm (AP), which was solely due to smaller systematic errors (table 2). For comparison, a 3 mm trigger threshold would lead to margins of 1.8 mm (LR), 5.2 mm (CC) and 2.8 mm (AP) with 0.6 (0–3) couch corrections per fraction while a 1 mm threshold would give margins of 1.3 mm (LR), 4.3 mm (CC) and 2.0 mm (AP) with 3.2 (0–6) couch corrections per fraction. If couch corrections can be performed without any time overhead, e.g. during gantry rotation between successive fields, one may use a 0 mm threshold. Such correction between all treatment fields would, however, only give a very modest further margin reduction to 1.2 mm (LR), 4.2 mm (CC) and 1.9 mm (AP). The remaining systematic error with this strategy (0.2 mm (LR), 0.3 mm (CC), 0.2 mm (AP)) is mainly due to baseline drift between the CBCT and the first treatment field.
An important aspect of gating and tracking guided by continuous internal motion monitoring is the possibility to reconstruct the dose delivered during radiotherapy. Such reconstruction of the actually delivered target dose is highly warranted for better understanding of dose–response relationships in liver SBRT and to guide the selection of the optimal treatment strategy for each patient (Jaffray et al 2010, Swaminath et al 2015).
As a limitation this study assumed the same starting time for each field in all simulated treatments as in the actual gated treatments in order to compare the motion adaptation strategies with the same intra-treatment motion. In reality, the four strategies would have different delivery times since treatments without motion adaptation and MLC tracking would be 5–8 min faster than gated treatments as discussed above. A reduction of the treatment time by 5–8 min would reduce the mean intrafraction baseline drift from the CBCT scan to the last field from ~3.8 mm to ~3.2 mm as estimated from figure 2(d) in Worm et al (2018). Furthermore, all motion mitigation strategies were investigated without use of abdominal compression although our clinic uses compression in most non-gated liver SBRT treatments to reduce the respiratory motion. Therefore, the motion and thus the motion-induced CTV dose deficiency are likely over-estimated in the simulated non-gated treatments. Another study limitation is that dose reconstruction was not performed for risk organs such as the duodenum since this motion may differ from the recorded liver motion. Furthermore, the use of an exhale planning CT scan rather than all phases of a 4DCT scan for tumor dose reconstruction in the non-gated scenarios can result in root-mean-square dose errors of approximately 1% in the high-dose region around the tumor (Poulsen et al 2015). Finally, all margin and target dose estimations were based on the transponder centroid motion only and did not include uncertainties arising from target definition, liver deformation, rotations or marker migration (Velec et al 2012, Bertholet et al 2016, 2017, Worm et al 2016).
An alternative option to manage intrafraction motion, which could not be simulated reliably from the free-breathing motion of this study, is gating with active breath-hold control. It should be noted that accurate internal intrafaction motion monitoring also is warranted for breath-hold treatments to ensure optimal treatment accuracy with reproducible tumor location at each breath hold (Zhong et al 2012, Lu et al 2018).
5. Conclusion
Four intrafraction motion adaptation strategies were compared for liver SBRT. Inter-field couch correction prevented gross target dose deficits without the requirement of real-time motion monitoring. However, gating and MLC tracking were more effective strategies that ensured high similarity between planned and delivered doses. Gating was slightly better than MLC tracking dosimetrically, but had a lower duty cycle and required several couch corrections to maintain the tumor exhale position inside the gating window.
Acknowledgments
This study was supported by grants from The Danish Cancer Society and Varian Medical Systems. The authors would like to thank Amir Movafeghi at the Radiation Application Research School, NSTRI, Tehran, for useful discussions.