- Email: [email protected]
Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver
Radiotherapy and Oncology xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Radiotherapy and Oncology journal homepage: www.thegreenjournal.com
Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver q Per Rugaard Poulsen a,b,⇑, Esben S. Worm a,c, Jørgen B.B. Petersen c, Cai Grau a,b, Walther Fledelius a,c, Morten Høyer a,b a
Department of Oncology, Aarhus University Hospital; b Institute of Clinical Medicine, Aarhus University; and c Department of Medical Physics, Aarhus University Hospital, Denmark
a r t i c l e
i n f o
Article history: Received 2 December 2013 Received in revised form 30 April 2014 Accepted 24 May 2014 Available online xxxx Keywords: Image-guided radiotherapy Intrafraction motion monitoring Dose reconstruction
a b s t r a c t Purpose: To use intrafraction kilovoltage (kV) imaging during liver stereotactic body radiotherapy (SBRT) delivered by volumetric modulated arc therapy (VMAT) to estimate the intra-treatment target motion and to reconstruct the delivered target dose. Methods: Six liver SBRT patients with 2–3 implanted gold markers received SBRT in three fractions of 18.75 Gy or 25 Gy. CTV-to-PTV margins of 5 mm in the axial plane and 10 mm in the cranio-caudal directions were applied. A VMAT plan was designed to give minimum target doses of 95% (CTV) and 67% (PTV). At each fraction, the 3D marker trajectory was estimated by fluoroscopic kV imaging throughout treatment delivery and used to reconstruct the actually delivered CTV dose. The reduction in D95 (minimum dose to 95% of the CTV) relative to the planned D95 was calculated. Results: The kV position estimation had mean root-mean-square errors of 0.36 mm and 0.47 mm parallel and perpendicular to the kV imager, respectively. Intrafraction motion caused a mean 3D target position error of 2.9 mm and a mean D95 reduction of 6.0%. The D95 reduction correlated with the mean 3D target position error during a fraction. Conclusions: Kilovoltage imaging for detailed motion monitoring with dose reconstruction of VMATbased liver SBRT was demonstrated for the first time showing large dosimetric impact of intrafraction tumor motion. Ó 2014 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology xxx (2014) xxx–xxx
Volumetric modulated arc therapy (VMAT) allows efficient delivery of highly conformal dose distributions in radiotherapy [1–3]. However, for thoracic and abdominal tumors respiratory motion may compromise the accuracy of the dose delivery. This is of particular concern for stereotactic body radiotherapy (SBRT) due to tightly conformed target dose distributions [4] that may be peaked in a central high dose region intended to coincide with the hypoxia-prone tumor center [5–10]. In liver SBRT, image-guidance often relies on implanted fiducial markers because liver tumors are difficult to visualize by in-room imaging. The implanted markers are typically used for patient setup [11], but the marker motion may also be monitored during treatment with stereoscopic kV imaging [12], MV portal imaging [13], or combined kV–MV imaging [14]. For a conventional linear
Abstract presentation: Presented in part at the 2nd ESTRO Forum, 19–23 April 2013, Geneva, Switzerland. ⇑ Corresponding author at: Aarhus University Hospital, Nr Brogade 44, 8000 Aarhus C, Denmark. E-mail address: [email protected] (P.R. Poulsen).
accelerator, stereoscopic kV imaging requires non-standard imagers while continuous localization with MV imaging may be unfeasible for VMAT where MLC motion sometimes hinders marker visibility [15]. These limitations have stimulated intra-treatment motion monitoring with a single kV imager, which is standard equipment for image-guided radiotherapy (IGRT) on many linear accelerators. Adamson and Wu used a single kV imager to estimate the three-dimensional (3D) prostate motion during intensity modulated radiotherapy by triangulation from different field directions, assuming static target positions between subsequent field deliveries [16]. This assumption reduces the temporal resolution of the estimated target motion and is invalid for thoracic and abdominal tumors. In a recent study, Ng et al. [17] acquired fluoroscopic kV images perpendicular to the treatment beam during prostate VMAT treatments and estimated the 3D target position for each individual kV image by a probability-based method [18]. Accurate intra-treatment 3D localization with the same frequency as used for the kV imaging (5–10 Hz) was obtained by this method, which was termed kilovoltage intrafraction monitoring (KIM) by the authors [17].
http://dx.doi.org/10.1016/j.radonc.2014.05.007 0167-8140/Ó 2014 Elsevier Ireland Ltd. All rights reserved.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
2
Motion and dose during VMAT liver SBRT
Several methods have been proposed to calculate the impact of intra-treatment motion on the delivered VMAT dose. For quasiperiodic respiratory motion, one approach is to split the VMAT plan into phase-specific sub-plans, calculate the dose of each sub-plan in the corresponding phase of a four-dimensional computed tomography (4DCT) scan, and accumulate the total dose using deformable image registration (DIR) [19,20]. Dose reconstruction with a phase specific patient anatomy can also be performed by Monte Carlo calculations using a transport grid with time dependent grid densities formed by interpolation of the 4DCT phases and mapping the dose back to a static reference grid by DIR [21]. While these approaches account for breathing induced anatomy deformations, they restrict the dose calculation to the motion range of the 4DCT scan, which may not cover the motion range of the treatment delivery. VMAT dose reconstruction beyond the 4DCT motion range can be obtained by constructing a time resolved dose matrix for the VMAT delivery to a static volume and trace the motion of each target voxel in this dose matrix while accumulating the voxel dose [22,23]. However, this approach neglects the 3D dose distribution modifications that occur even with rigid tumor shifts and may be quite substantial for lung tumors. This limitation can be overcome by modeling the tumor motion during VMAT as multiple isocenter shifts [24]. Phantom studies have demonstrated that this method results in accurate dose reconstruction for dynamic treatments delivered to a rigidly moving water density target embedded in lung density material [24], but the method has not yet been applied clinically for VMAT treatments of respiratory moving tumors. The purpose of this study was to combine the novel methods of KIM [17] and dynamic dose reconstruction [24] for VMAT-based liver SBRT treatments, in order to perform a detailed investigation of the dosimetric impact of intrafraction motion during actual clinical liver SBRT treatments. Methods and materials Method overview Fig. 1 summarizes the methods of this study. Fluoroscopic kV images acquired during VMAT-based liver SBRT (Fig. 1a) were used to estimate the 3D trajectory of an implanted gold marker [17,18] (Fig. 1b, c), which was in turn used to reconstruct the actually delivered target dose distribution [24] (Fig. 1d). Patients, planning, and treatments This study includes six patients with liver metastases who received SBRT delivered by VMAT in three fractions between March 2011 and February 2013, following our standard SBRT treatment protocol. Guided by ultrasound, each patient had 2–3 cylindrical gold markers (1 mm 3 mm) implanted close to the metastases. The clinical target volume (CTV) was delineated in the mid-ventilation phase of a 4DCT scan with 3 mm slice thickness. The planning
target volume (PTV) was formed by adding 5 mm margins in the left–right (LR) and anterior–posterior (AP) directions and 10 mm in the cranio-caudal (CC) direction. Using the Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA, USA) a 6MV VMAT plan with 5–6 arcs was designed to give a CTV mean dose of 100% and minimum target doses of 95% (CTV) and 67% (PTV), i.e. a non-uniform PTV dose distribution with 50% higher dose in the central part [5–8]. Following a risk-adapted strategy, fraction doses of 12.56 Gy or 16.75 Gy were prescribed to the 67% isodose surface (tightly enclosing the PTV surface) resulting in 100% dose levels of 18.75 Gy or 25 Gy. The arc fields spanned between 134° and 234° and were delivered with a maximum dose rate of 600 MU/min. The degree of VMAT intensity modulation can be characterized by the number of monitor units (MU) per Gy (Table 1) or the mean MLC aperture area relative to the total area covered by the arcs, which ranged from 15% to 47% for the six VMAT plans (average 27%). The table rotation was 15° for four out of six arcs for one patient (Patient 2) and 0° for all other arcs. A Stereotactic Body Frame (Elekta, Crawley, UK) or an in-house modified breast board was used for immobilization. Abdominal compression was used for all patients. Table 1 summarizes relevant patient and image data. The treatments were delivered using five different Trilogy accelerators equipped with an MV PortalVision AS500 or AS1000 portal imager and an On-Board Imager (OBI) (Varian Medical Systems). Prior to treatment, a cone-beam CT scan was used for markerbased patient setup. Continuous portal images (12.8 Hz for AS500, 7.5 Hz for AS1000, 160 cm source-imager-distance (SID)) and orthogonal kV images (5.0 Hz, 125 kV, 80 mA, 11–19 ms, fullfan bow-tie filter, 180 cm SID) were acquired throughout the treatment delivery. The kV field size (Table 1) was chosen as the patient specific minimum size that covered all markers plus a 1.5–2 cm margin as seen from all kV imaging angles of the arc field. The effective dose from intra-treatment kV imaging was coarsely estimated to be 3–6 mSv per fraction from a CBCT scan mode with similar settings (11.0 Hz, 125 kV, 80 mA, 25 ms, full-fan bow-tie filter) by scaling the CBCT dose (9.1 mSv) with the number of images, CC field size, and exposure duration (Table 1). The image and motion data of one patient were previously reported as part of a pilot study on image-based intrafraction motion monitoring (Patient 1, who is identical to Patient 6 in Ref. [14]) while image data for the remaining five patients and all dose reconstruction data and KIM accuracy data are new in this study. Cross-scattering of the MV treatment beam onto the kV imager degraded the quality of the intra-treatment kV images. As a coarse indicator of the VMAT field size and thus the amount of MV irradiation hitting the patient, Table 1 shows the average jaw size for each treatment plan. Since the kV image quality also depended on the kV beam path length through the patient, Table 1 presents the longest kV water equivalent path length (WEPL) that passed through the monitored gold marker. Intra-treatment kV imaging was not attempted for three other liver SBRT patients treated with VMAT in the same period as the
Fig. 1. Method overview.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
3
P.R. Poulsen et al. / Radiotherapy and Oncology xxx (2014) xxx–xxx
Table 1 Patient and image data, including mean CTV dose, number of metastases, effective dose per fraction by intra-treatment kV imaging (E), maximum water equivalent path length of the kV imaging rays through the selected marker (Max kV WEPL), and the kV field size and MV jaw field size averaged over all arc fields of the VMAT plan. Patient
Sex
Age (years)
Metastases
CTV dose (Gy)
VMAT fields
MU/Gy
kV images per field
kV field size (cm x cm)
E (mSv)
Max kV WEPL (cm)
MV jaws (cm cm)
MV imager
1 2 3 4 5 6
F M F M F M
60 79 75 73 68 56
6 4 2 1 1 1
3 18.75 3 18.75 3 25 3 25 3 25 3 25
5 6 5 5 5 5
200 319 167 186 171 161
415 532 439 503 466 434
12.5 3.6 7.5 5.7 15.2 4.2 6.8 2.7 9.2 4.8 8 5.5
3 6 3 3 3 4
28.3 34.8 31.3 37.1 31.8 35.9
17.9 14.8 14.0 14.0 12.1 10.7 8.0 6.9 7.9 5.7 9.6 8.6
AS1000 AS1000 AS1000 AS500 AS1000 AS500
six included patients because a combination of large field sizes and long kV WEPL indicated the possibility of poor intra-treatment kV image quality.
Intra-treatment 3D motion After the treatments, an automatic in-house developed segmentation algorithm was used to determine the position of the gold marker closest to the isocenter in all kV images with a resolution of 1 pixel (0.22 mm) [25]. The auto-segmentation was manually inspected in all kV images and corrected if the segmented position deviated more than 3 pixels (0.7 mm) from the marker center. The marker segmentation provided the projected 2D motion of the selected marker in the rotating kV imager plane (Fig. 1b). After correction for gantry and image system sag [26] the 2D motion was used to estimate the 3D marker motion in the patient (Fig. 1c) by a probability based method [18,27] that uses a fitted 3D Gaussian probability density function to estimate the unknown marker position perpendicular to the kV imager (and thus the full 3D position) for each kV image. The group systematic error M in the intra-treatment marker position relative to the planned position and the standard deviation of the systematic error R and random error r were calculated following Ref. [28]. For some of the non-coplanar arcs of Patient 2, the marker was not visible by eye in the kV images in one or two periods of 1–2 s duration because of excessive MV cross-scatter onto the kV imager. The internal 3D marker motion during these short periods was reconstructed from the motion of an abdominal marker block (RPM, Varian Medical System) by use of a linear motion correlation model. The correlation model was established using the internal 3D marker motion of the segmentable kV images of the arc field and had an estimated accuracy of 1 mm or less. The accuracy of the kV-based 3D marker position estimation was determined from the continuous MV images as follows: For the four patients imaged with an AS1000 portal imager (Table 1), the marker was manually segmented in all MV images with marker visibility at one treatment fraction. The kV-estimated 3D trajectory was projected onto the portal imager and the kV position estimation error was calculated as the difference (scaled to isocenter distance) between the projected positions and the segmented marker positions in the MV imager (Fig. 1, bottom). This validation of the kV position estimation was not performed for the two patients imaged with an AS500 portal imager because the marker contrast was too poor for reliable segmentation. The probability based 3D position estimation method not only provided the target position for each kV image, but also the rootmean-square (rms) error of the target position estimation [27,29] in the unknown direction perpendicular to the kV imager. The rms over all kV images of this rms estimation error was calculated for each patient. Note that this error (unlike the MV determined error above) only contains contributions from the probability based 3D position estimation method and not from segmentation uncertainties or image system sag.
Dose reconstruction For all fractions, the 3D marker motion was used to reconstruct the delivered target dose by an experimentally validated method [24,30] that models the motion of a rigid target as multiple isocenter shifts in an in-house built software program (Matlab) and utilizes the clinical treatment planning system for the dose calculation (Eclipse 11.0, Anisotropic Analytical Algorithm (AAA), Varian Medical Systems). The calculations assumed that the MLC positions as function of gantry angle were equal to the planned MLC positions. A 1.5 mm bin width was used, i.e. all target positions within a 1.5 1.5 1.5 mm3 cube were given the same isocenter shift. For each fraction and treatment course, the reduction of the CTV mean dose (DDmean) and of the minimum dose to 95% of the CTV (DD95) relative to the treatment plan was calculated. Effects of anatomy changes on reconstructed dose The motion including dose reconstruction accounts for rigid target motion, interplay effects, and physical path length changes, but neglects target deformations, radiological path length changes, and target shifts relative to the gold marker [24]. While all dose calculations were performed by applying the isocenter shifts in the midventilation phase of the 4DCT scan, breathing induced anatomy deformation could have been included by distributing the dose calculations among all 4DCT phases [19,20]. However, this would restrict the dose calculations to the target shifts observed in the 4DCT scan, whereas a much wider range of shifts may occur at treatment [31]. The maximum erroneous effect of performing all dose reconstructions in the mid-ventilation phase was estimated as follows for each patient: First, the original treatment plan was recalculated in the full inhale and full exhale phases, which provided reference dose distributions in these extreme phases including the effects of anatomical changes. Next, the plan was recalculated in the mid-ventilation phase, but with the isocenter shifted corresponding to the marker shift between the mid-ventilation phase the extreme phases, which mimicked the dose reconstruction method in these phases. The root-mean-square (rms) dose difference between the dose distributions of the two calculations (in the extreme 4DCT phase and the mid-ventilation phase, respectively) was calculated by in-house developed software as a measure of the dose reconstruction error caused by the most extreme anatomical changes observed in the 4DCT scans. While this analysis quantified the maximum errors in the spatial dose distribution within the patient volume it did not assess errors in the reconstructed CTV dose distribution stemming from deformations of the CTV or shifts of the CTV relative to the monitored marker. Results Intra-treatment 3D motion Fig. 2 shows an example of the 3D intra-treatment liver tumor motion determined by KIM at a treatment fraction with relatively
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
4
Motion and dose during VMAT liver SBRT
regular breathing. At this fraction, the intra-treatment peak-topeak motion range was 5.4 mm (LR), 14.2 mm (CC), and 8.4 mm (AP). A gradual drift during the treatment resulted in relatively large baseline motion ranges (i.e. range of intra-field mean positions) of 2.1 mm (LR), 2.7 mm (CC), and 1.8 mm (AP). The mean position error (i.e. the patient-specific systematic error) was 3.5 mm (LR), 2.5 mm (CC), 0.0 mm (AP), and 4.3 mm (3D). The relatively large mean position error was caused by an intrafraction tumor shift between the setup CBCT imaging and treatment start. Table 2 presents the mean motion ranges and position errors for all patients. The average mean 3D position error was 2.9 mm for single fractions and 2.1 mm for treatment courses. The largest motion ranges during any treatment fraction was 8.7 mm (LR), 35.4 mm (CC), 9.7 mm (AP), and 35.7 (3D). The population based group mean error and the standard deviations of systematic and random errors in the LR, CC, and AP directions were: M = 1.0 mm, 1.1 mm, and 0.7 mm; R = 1.1 mm, 1.7 mm, and 1.0 mm; and r = 1.2 mm, 3.9 mm, and 1.6 mm, respectively. The gold marker was visible within the MLC aperture in 22% of the validation MV images (Table 2). The high rate of marker occlusion by the MLC was partly caused by marker implantation outside the CTV in order to avoid tumor cell spread. The mean [and maximum] rms position estimation error of the KIM method, as determined by the MV images, was 0.36 mm [0.61 mm] parallel to the kV imager and 0.47 mm [0.81 mm] perpendicular to the kV imager (Fig. 2, bottom graphs, Table 2). The probability based
position estimation method itself estimated an rms position estimation error perpendicular to the kV imager of 0.34 mm when averaged over all 18 fractions (Table 2). Dose reconstruction Fig. 3 presents the reconstructed doses and resulting CTV dosevolume-histograms for Patient 1. Compared to the planned dose distribution the delivered doses were blurred due to motion, shifted due to target position errors, and redistributed due to interplay effects. The average reduction in mean CTV dose at treatment was 2.2% (Table 2). The CTV DD95 was in mean 6.0% for single fractions and 5.3% for treatment courses (Table 2), and it correlated with the mean 3D position error (p < 0.001, Pearson, Fig. 4). In some cases, DD95 was however substantial even though the mean position error was modest (Fig. 4, points labeled Patient 1 and Patient 2). The reason for this was large random errors caused either by large mean position errors in different directions at the three treatment fractions (Patient 1 course dose, see Fig. 3) or by large respiratory motion (Patient 2 fraction and course doses, see Table 2). Effects of anatomy changes on reconstructed dose The mean rms difference between the dose distributions reconstructed with isocenter shifts in the mid-ventilation phase and the
Fig. 2. Three upper graphs: Intra-treatment target motion (black) relative to the planned position (red line) in left–right (LR), cranio-caudal (CC), and anterior–posterior (AP) direction as estimated by kilovoltage intrafraction motion monitoring during delivery of five VMAT fields for Patient 5 at fraction 2. Two lower graphs: Projection of the estimated 3D positions onto the MV imager (black) compared with the actual position measured in the MV images (red). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Table 2 Mean over three fractions of the intra-treatment marker motion ranges and magnitude of mean target position errors. Root-mean-square errors (rmse) of kilovoltage intrafraction monitoring (KIM) measured by MV imaging (parallel and perpendicular to the kV imager) and estimated by the KIM method itself (perpendicular to the kV imager). Reduction in CTV D95 and CTV mean dose for each treatment course. Pt
1 2 3 4 5 6 All
Motion range (mm) Total LR/CC/AP
Baseline LR/CC/AP
3.7/12.7/2.4 2.9/27.1/8.2 5.4/8.8/5.2 5.6/21.1/8.7 4.7/12.1/7.7 6.2/15.6/6.5 4.8/16.2/6.4
0.8/1.1/0.4 0.6/2.1/0.8 1.9/1.1/1.0 0.5/1.5/0.7 1.2/1.5/1.0 0.5/1.4/0.7 0.9/1.4/0.8
Magnitude of error (mm)
MV images Total
% with marker
Parallel to kV mean [max]
Perpendicular to kV mean [max]
2958 4586 3143 – 3355 – 14042
23 3 32 – 40 – 22
0.43 0.41 0.30 – 0.30 – 0.36
0.41 0.41 0.38 – 0.67 – 0.47
LR/CC/AP 1.0/3.0/1.3 1.3/1.0/0.7 1.3/0.6/0.4 0.5/3.0/1.5 3.1/3.4/1.8 1.1/0.8/1.1 1.4/2.0/1.1
KIM rmse in MV images (mm)
[0.50] [0.61] [0.35] [0.48] [0.61]
[0.49] [0.60] [0.40] [0.81] [0.81]
KIM rmse (mm)
CTV DD95 (%)
CTV DDmean (%)
0.28 0.28 0.38 0.33 0.42 0.36 0.34
4.0 9.5 0.3 6.0 11.1 1.0 5.3
1.6 5.4 0.8 2.2 3.0 0.2 2.2
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
P.R. Poulsen et al. / Radiotherapy and Oncology xxx (2014) xxx–xxx
5
Fig. 3. Left: Planned and actually delivered dose distributions shown as >95% dose color wash in a coronal plane with the GTV (red contour) and PTV (blue) for Patient 1. Right: Corresponding CTV dose volume histograms. DD95 denotes the reduction in the minimum dose to 95% of the CTV at treatment. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 4. 3D target position error versus reduction in CTV D95 for individual treatment fractions (small black circles) and treatment courses (gray).
reference dose distributions in the extreme 4DCT phases was 0.55% and 0.64% of the prescribed dose when including all voxels with doses larger than 10% dose and 90% dose, respectively. These numbers represent errors in the reconstructed dose distribution caused by large anatomical changes in general and in high dose regions, respectively. The rms dose difference correlated with the 3D target shift |Dr| between the mid-ventilation and extreme phases (p < 0.0001, Pearson) and was to first order equal to 0.10%/mm |Dr| (in general) and 0.16%/mm |Dr| (high dose regions). The 3D tumor motion range was considerably larger during treatment delivery (mean: 17.3 mm; range: 7.3–35.7 mm) than in the planning 4DCT scan (mean: 7.6 mm; range: 4.0–14.7 mm) showing that the 4DCT scans failed to capture the range of anatomy changes occurring at treatment. Linear extrapolation of the rms dose difference in the 4DCT scans to the most extreme target shift observed during treatment (|Dr|max = 27.5 mm, Patient 2, fraction 3) yielded rms dose calculation errors of 2.7% (in general) and 4.4% (high dose regions) for this extreme position and of 0.57% (in general) and 0.92% (high dose regions) when time averaged over all target positions at this particular fraction. Discussion For the first time, kilovoltage intrafraction motion monitoring with dose reconstruction was demonstrated for respiratory moving targets. At 18 liver SBRT treatment fractions, the 3D intra-treatment target motion was accurately estimated throughout VMAT
delivery using the standard IGRT equipment of a linear accelerator. Validation MV imaging demonstrated sub-millimeter localization accuracy for four treatment fractions, while the inherent error estimation of the single-imager localization method indicated similar accuracy for the fractions not monitored by MV imaging. The observed motion was in accordance with previous studies of liver tumor motion [12,14,31,32]. The baseline drift in the example in Fig. 2 agrees with MR observations of systematic intrafraction drifts of the liver [33], but it was not a general motion trend in the current study. Despite CBCT guided patient setup large 3D mean position errors occasionally occurred (up to 6.6 mm) leading to substantial deviations between planned and delivered CTV doses (Fig. 4). Post-treatment review of the online CBCT setup for the treatments with largest position errors showed that, within the uncertainties of marker-based CBCT setup [11], the online marker-based registration was performed correctly. The occasionally large position errors primarily resulted from baseline shifts occurring between IGRT imaging and treatment completion. The correlation between DD95 and the 3D mean position error suggests that a relatively simple correction strategy involving inter-field couch adjustments in case of baseline shifts may give substantial dosimetric improvements. However, this correction strategy would not mitigate dosimetric errors caused by large respiratory motion (e.g. Patient 2 in Fig. 4). Instead, KIM may be combined with real-time marker segmentation and used for real-time motion adaptation during VMAT delivery [34–35]. Despite large CTV D95 reductions, all patients in this study received the prescribed dose since the dose was prescribed to the 67% dose level, whereas the minimum CTV dose at any fraction [course] was 71% [76%]. The large CTV D95 reductions were caused by CTV motion relative to the central high dose region of the PTV. The CTV dose would be less susceptible to intrafraction motion if a uniform PTV dose were prescribed rather than a PTV dose that gradually decreased from 95% to 67% outside the CTV [5–10]. However, in order to be isotoxic a uniform PTV dose would mean a lower prescribed dose to the hypoxia-prone tumor center than with the peaked PTV dose prescription. Because of large uncertainties in current modeling of the tumor control probability and normal tissue complication probability, in particular for non-uniform dose distributions, it is not obvious which prescription strategy that would be optimal. While the presented KIM motion data and dose reconstruction method may be used in combination with biological modeling to investigate dose prescription optimization under consideration of intrafraction tumor motion, this subject is beyond the scope of the current study. A limitation of the KIM method is the additional imaging dose, which was comparable to the dose of a CBCT scan in this study (Table 1). The number of single images acquired per second (five)
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
6
Motion and dose during VMAT liver SBRT
was one order of magnitude lower than with the RTRT system [12], which uses stereoscopic imaging at 30 Hz for internal marker motion monitoring. Real-time tracking of lung tumors has been performed with the Vero4DRT system based on an external-internal motion model that was established prior to treatment by stereoscopic imaging at 6.25 Hz or 12.5 Hz for 40 s [36] and monitored by 1 Hz stereoscopic imaging during 6–7 field deliveries of 32 s duration [37]. It gives up to 1448 single kV images plus additional imaging for 1–2 model updates at each fraction [36]. The current study obtained direct internal motion monitoring with 2100–3200 kV images per fraction (Table 1). The direct internal (surrogate) monitoring allows use of KIM for targets without external motion correlation such as the prostate [17]. The current KIM implementation suffers from excessive MV cross-scatter onto the kV imager, which made the marker invisible in some kV images for one patient, challenged robust automatic marker segmentation in general, and caused us to omit kV imaging for three patients treated with VMAT-based liver SBRT in the same period. Improved kV imager read-out that includes only the kV exposure period (11–19 ms) rather than the entire period between kV images (200 ms) would reduce the duration of MV scatter exposure of the kV images by one order of magnitude. In this study, the reconstructed dose included interplay effects as well as shifts and blurring of the target dose caused by systematic and random intrafraction motion. Separation of the dose effects into these components were beyond the scope of this study. However, we did not observe any noticeable interplay effects such as cold spots within high dose regions. This is in agreement with recent studies of respiration induced interplay effects during VMAT SBRT [19,20,23]. Only negligible interplay effects may be expected as the delivery time at each treatment fraction covered more than 100 breathing cycles. The dose reconstruction in this study accounted for target motion, interplay effects, and physical path length changes, it did not account for breathing induced anatomy deformations since all calculations were performed in the mid-ventilation phase of the planning 4DCT scan. This simplification caused estimated rms errors in the dose distribution in the patient of up to 1% for the patient with largest intra-treatment motion. However, within the reconstructed dose distribution we assumed that the CTV kept the same shape and position relative to the gold marker as in the mid-ventilation planning CT phase. Based on deformable registration between exhale and inhale CT volumes Velec et al. studied dose accumulation over the breathing cycle in liver SBRT and found that, compared to deformable registration, simple rigid liver-to-liver registration resulted in up to 8% discrepancies in the accumulated minimum tumor dose [38]. However, in their rigid registration dose accumulation, the authors assumed that the tumor motion was identical to the liver center-of-mass motion, which was obviously not a valid assumption since deformable registration revealed differences between tumor and liver center-of-mass motion of up to 6 mm (LR), 10 mm (CC), and 7 mm (AP) for these patients [38]. In the current study, we used an implanted gold marker (and not the liver center-of-mass) as a surrogate for the CTV position, which is a much more reasonable assumption [39]. Therefore, the lack of deformation modeling will have a much smaller impact on the accumulated tumor dose than in Ref. [38]. In conclusion, kilovoltage intrafraction motion monitoring with dose reconstruction for VMAT-based liver SBRT was demonstrated for the first time revealing large dosimetric impact of intrafraction motion. Since the motion measurements and dose reconstruction rely on standard equipment and software for IGRT and treatment planning, respectively, the methods have the potential for widespread use. The improved reporting of the actually delivered CTV dose could lead to better understanding of clinical dose–response relationships in radiation oncology.
Conflict of interest This work was partially supported by a research grant from Varian Medical Systems, Inc. (Palo Alto, CA). P.R. Poulsen receives royalties from a patent related to this work. Acknowledgments This work was supported The Danish Cancer Society, CIRRO – The Lundbeck Foundation Center for Interventional Research in Radiation Oncology, and Varian Medical Systems, Inc., Palo Alto, CA. References [1] Yu CX. Intensity-modulated arc therapy with dynamic multileaf collimation: an alternative to tomotherapy. Phys Med Biol 1995;40:1435–49. [2] Otto K. Volumetric modulated arc therapy: IMRT in a single gantry arc. Med Phys 2008;35:310–7. [3] Wang C, Luan S, Tang G, Chen DZ, Earl MA, Yu CX. Arc-modulated radiation therapy (AMRT): a single-arc form of intensity-modulated arc therapy. Phys Med Biol 2008;53:6291–303. [4] Benedict SH, Yenice KM, Followill D, et al. Stereotactic body radiation therapy: the report of AAPM Task Group 101. Med Phys 2010;37:4078–101. [5] Lax I, Blomgren H, Näslund I, Svanström R. Stereotactic radiotherapy of malignancies in the abdomen. Methodological aspects. Acta Oncol 1994;33:677–83. [6] Lax I. Target dose versus extra target dose in stereotactic radiosurgery. Acta Oncol 1993;32:453–7. [7] Lax I, Panettieri V, Wennberg B, et al. Dose distribution in SBRT of lung tumors: comparison between two different treatment planning algorithms and MonteCarlo simulation including breathing motion. Acta Oncol 2006;45:978–88. [8] Hoyer M, Roed H, Traberg Hansen A, et al. Phase II study on stereotactic body radiotherapy of colorectal metastases. Acta Oncol 2006;45:823–30. [9] de Pooter JA, Wunderink W, Mendez Romero A, et al. PTV dose prescription strategies for SBRT of metastatic liver tumours. Radiother Oncol 2007;85:260–6. [10] Molinelli S, de Pooter J, Mendez Romero A, et al. Simultaneous tumour dose escalation and liver sparing in Stereotactic Body Radiation Therapy (SBRT) for liver tumours due to CTV-to-PTV margin reduction. Radiother Oncol 2008;87:432–8. [11] Worm ES, Høyer M, Fledelius W, Nielsen JE, Larsen LP, Poulsen PR. On-line use of three-dimensional marker trajectory estimation from cone-beam computed tomography projections for precise setup in radiotherapy for targets with respiratory motion. Int J Radiat Oncol Biol Phys 2012;83:e145–51. [12] Shirato H, Seppenwoolde Y, Kitamura K, et al. Intrafractional tumor motion: lung and liver. Semin Radiat Oncol 2004;14:10–8. [13] Berbeco RI, Neicu T, Rietzel E, et al. A technique for respiratory-gated radiotherapy treatment verification with an EPID in cine mode. Phys Med Biol 2005;50:3669–79. [14] Worm ES, Høyer M, Fledelius W, Poulsen PR. Three-dimensional, timeresolved, intrafraction motion monitoring throughout stereotactic liver radiation therapy on a conventional linear accelerator. Int J Radiat Oncol Biol Phys 2013;86:190–7. [15] Azcona JD, Li R, Mok E, et al. Development and clinical evaluation of automatic fiducial detection for tumor tracking in cine megavoltage images during volumetric modulated arc therapy. Med Phys 2013;40:031708. [16] Adamson J, Wu Q. Prostate intrafraction motion evaluation using kV fluoroscopy during treatment delivery: a feasibility and accuracy study. Med Phys 2008;35:1793–806. [17] Ng JA, Booth JT, Poulsen PR, et al. Kilovoltage intrafraction monitoring for prostate intensity modulated arc therapy: first clinical results. Int J Radiat Oncol Biol Phys 2012;84:e655–61. [18] Poulsen PR, Cho B, Keall PJ. A method to estimate mean position, motion magnitude, motion correlation, and trajectory of a tumor from cone-beam CT projections for image-guided radiotherapy. Int J Radiat Oncol Biol Phys 2008;72:1587–96. [19] Rao M, Wu J, Cao D, et al. Dosimetric impact of breathing motion in lung stereotactic body radiotherapy treatment using image-modulated radiotherapy and volumetric modulated arc therapy. Int J Radiat Oncol Biol Phys 2012;83:e251–6. [20] Li X, Yang Y, Li T, et al. Dosimetric effect of respiratory motion on volumetricmodulated arc therapy-based lung SBRT treatment delivered by TrueBeam machine with flattening filter-free beam. J Appl Clin Med Phys 2013;14:195–204. [21] Belec J, Clark BG. Monte Carlo calculation of VMAT and helical tomotherapy dose distributions for lung stereotactic treatments with intra-fraction motion. Phys Med Biol 2013;58:2807–21. [22] Feygelman V, Stambaugh C, Zhang G, et al. Motion as a perturbation: measurement-guided dose estimates to moving patient voxels during modulated arc deliveries. Med Phys 2013;40:021708.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
P.R. Poulsen et al. / Radiotherapy and Oncology xxx (2014) xxx–xxx [23] Stambaugh C, Nelms BE, Dilling T, et al. Experimentally studied dynamic dose interplay does not meaningfully affect target dose in VMAT SBRT lung treatments. Med Phys 2013;40:091710. [24] Poulsen PR, Schmidt ML, Keall P, Worm ES, Fledelius W, Hoffmann L. A method of dose reconstruction for moving targets compatible with dynamic treatments. Med Phys 2012;39:6237–46. [25] Fledelius W, Worm E, Elstrom UV, et al. Robust automatic segmentation of multiple implanted cylindrical gold fiducial markers in cone-beam CT projections. Med Phys 2011;38:6351. [26] Cho B, Poulsen PR, Sloutsky A, Sawant A, Keall P. First demonstration of combined kV/MV image-guided real-time DMLC target tracking. Int J Radiat Oncol Biol Phys 2009;74:859–67. [27] Poulsen PR, Cho B, Keall PJ. Real-time prostate trajectory estimation with a single imager in arc radiotherapy: a simulation study. Phys Med Biol 2009;54:4019–35. [28] van Herk M. Errors and margins in radiotherapy. Semin Radiat Oncol 2004;14:52–64. [29] Poulsen PR, Cho B, Keall PJ. Prediction of position estimation errors for 3D target trajectories estimated from cone-beam CT projections. In: Proceedings of XVIth international conference on the use of computers in radiation therapy, Amsterdam, 31/5/2010–3/6/2010. [30] Ravkilde T, Keall PJ, Grau C, Høyer M, Poulsen PR. Time-resolved dose reconstruction by motion encoding of volumetric modulated arc therapy fields delivered with and without dynamic multi-leaf collimator tracking. Acta Oncol 2013;52:1497–503. [31] Worm ES, Høyer M, Fledelius W, Hansen AT, Poulsen PR. Variations in magnitude and directionality of respiratory target motion throughout full
[32]
[33]
[34]
[35] [36]
[37]
[38]
[39]
7
treatment courses of stereotactic body radiotherapy for tumors in the liver. Acta Oncol 2013;52:1437–44. Gabrys D, Kulik R, Trela K, Slosarek K. Dosimetric comparison of liver tumour radiotherapy in all respiratory phases and in one phase using 4DCT. Radiother Oncol 2011;100:360–4. von Siebenthal M, Székely G, Lomax AJ, Cattin PC. Systematic errors in respiratory gating due to intrafraction deformations of the liver. Med Phys 2007;34:3620–9. Poulsen PR, Fledelius W, Cho B, Keall P. Image-based dynamic multileaf collimator tracking of moving targets during intensity-modulated arc therapy. Int J Radiat Oncol Biol Phys 2012;83:e265–71. Falk M, af Rosenschöld PM, Keall P, et al. Real-time dynamic MLC tracking for inversely optimized arc radiotherapy. Radiother Oncol 2010;94:218–23. Akimoto M, Nakamura M, Mukumoto N, et al. Predictive uncertainty in infrared marker-based dynamic tumor tracking with Vero4DRT. Med Phys 2013;40:091705. Nakamura M, Akimoto M, Mukumoto N, et al. Influence of predictive modelling duration on the predictive accuracy of IR-marker-based dynamic tumour tracking. Radiother Oncol 2013;108:S320. Velec M, Moseley JL, Eccles CL, et al. Effect of breathing motion on radiotherapy dose accumulation in the abdomen using deformable registration. Int J Radiat Oncol Biol Phys 2011;80:265–72. Seppenwoolde Y, Wunderink W, Wunderink-van Veen SR, Storchi P, Mendes Romero A, Heijmen BJ. Treatment precision of image-guided liver SBRT using implanted fiducial markers depends on marker-tumour distance. Phys Med Biol 2011;56:5445–68.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
Contents lists available at ScienceDirect
Radiotherapy and Oncology journal homepage: www.thegreenjournal.com
Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver q Per Rugaard Poulsen a,b,⇑, Esben S. Worm a,c, Jørgen B.B. Petersen c, Cai Grau a,b, Walther Fledelius a,c, Morten Høyer a,b a
Department of Oncology, Aarhus University Hospital; b Institute of Clinical Medicine, Aarhus University; and c Department of Medical Physics, Aarhus University Hospital, Denmark
a r t i c l e
i n f o
Article history: Received 2 December 2013 Received in revised form 30 April 2014 Accepted 24 May 2014 Available online xxxx Keywords: Image-guided radiotherapy Intrafraction motion monitoring Dose reconstruction
a b s t r a c t Purpose: To use intrafraction kilovoltage (kV) imaging during liver stereotactic body radiotherapy (SBRT) delivered by volumetric modulated arc therapy (VMAT) to estimate the intra-treatment target motion and to reconstruct the delivered target dose. Methods: Six liver SBRT patients with 2–3 implanted gold markers received SBRT in three fractions of 18.75 Gy or 25 Gy. CTV-to-PTV margins of 5 mm in the axial plane and 10 mm in the cranio-caudal directions were applied. A VMAT plan was designed to give minimum target doses of 95% (CTV) and 67% (PTV). At each fraction, the 3D marker trajectory was estimated by fluoroscopic kV imaging throughout treatment delivery and used to reconstruct the actually delivered CTV dose. The reduction in D95 (minimum dose to 95% of the CTV) relative to the planned D95 was calculated. Results: The kV position estimation had mean root-mean-square errors of 0.36 mm and 0.47 mm parallel and perpendicular to the kV imager, respectively. Intrafraction motion caused a mean 3D target position error of 2.9 mm and a mean D95 reduction of 6.0%. The D95 reduction correlated with the mean 3D target position error during a fraction. Conclusions: Kilovoltage imaging for detailed motion monitoring with dose reconstruction of VMATbased liver SBRT was demonstrated for the first time showing large dosimetric impact of intrafraction tumor motion. Ó 2014 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology xxx (2014) xxx–xxx
Volumetric modulated arc therapy (VMAT) allows efficient delivery of highly conformal dose distributions in radiotherapy [1–3]. However, for thoracic and abdominal tumors respiratory motion may compromise the accuracy of the dose delivery. This is of particular concern for stereotactic body radiotherapy (SBRT) due to tightly conformed target dose distributions [4] that may be peaked in a central high dose region intended to coincide with the hypoxia-prone tumor center [5–10]. In liver SBRT, image-guidance often relies on implanted fiducial markers because liver tumors are difficult to visualize by in-room imaging. The implanted markers are typically used for patient setup [11], but the marker motion may also be monitored during treatment with stereoscopic kV imaging [12], MV portal imaging [13], or combined kV–MV imaging [14]. For a conventional linear
Abstract presentation: Presented in part at the 2nd ESTRO Forum, 19–23 April 2013, Geneva, Switzerland. ⇑ Corresponding author at: Aarhus University Hospital, Nr Brogade 44, 8000 Aarhus C, Denmark. E-mail address: [email protected] (P.R. Poulsen).
accelerator, stereoscopic kV imaging requires non-standard imagers while continuous localization with MV imaging may be unfeasible for VMAT where MLC motion sometimes hinders marker visibility [15]. These limitations have stimulated intra-treatment motion monitoring with a single kV imager, which is standard equipment for image-guided radiotherapy (IGRT) on many linear accelerators. Adamson and Wu used a single kV imager to estimate the three-dimensional (3D) prostate motion during intensity modulated radiotherapy by triangulation from different field directions, assuming static target positions between subsequent field deliveries [16]. This assumption reduces the temporal resolution of the estimated target motion and is invalid for thoracic and abdominal tumors. In a recent study, Ng et al. [17] acquired fluoroscopic kV images perpendicular to the treatment beam during prostate VMAT treatments and estimated the 3D target position for each individual kV image by a probability-based method [18]. Accurate intra-treatment 3D localization with the same frequency as used for the kV imaging (5–10 Hz) was obtained by this method, which was termed kilovoltage intrafraction monitoring (KIM) by the authors [17].
http://dx.doi.org/10.1016/j.radonc.2014.05.007 0167-8140/Ó 2014 Elsevier Ireland Ltd. All rights reserved.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
2
Motion and dose during VMAT liver SBRT
Several methods have been proposed to calculate the impact of intra-treatment motion on the delivered VMAT dose. For quasiperiodic respiratory motion, one approach is to split the VMAT plan into phase-specific sub-plans, calculate the dose of each sub-plan in the corresponding phase of a four-dimensional computed tomography (4DCT) scan, and accumulate the total dose using deformable image registration (DIR) [19,20]. Dose reconstruction with a phase specific patient anatomy can also be performed by Monte Carlo calculations using a transport grid with time dependent grid densities formed by interpolation of the 4DCT phases and mapping the dose back to a static reference grid by DIR [21]. While these approaches account for breathing induced anatomy deformations, they restrict the dose calculation to the motion range of the 4DCT scan, which may not cover the motion range of the treatment delivery. VMAT dose reconstruction beyond the 4DCT motion range can be obtained by constructing a time resolved dose matrix for the VMAT delivery to a static volume and trace the motion of each target voxel in this dose matrix while accumulating the voxel dose [22,23]. However, this approach neglects the 3D dose distribution modifications that occur even with rigid tumor shifts and may be quite substantial for lung tumors. This limitation can be overcome by modeling the tumor motion during VMAT as multiple isocenter shifts [24]. Phantom studies have demonstrated that this method results in accurate dose reconstruction for dynamic treatments delivered to a rigidly moving water density target embedded in lung density material [24], but the method has not yet been applied clinically for VMAT treatments of respiratory moving tumors. The purpose of this study was to combine the novel methods of KIM [17] and dynamic dose reconstruction [24] for VMAT-based liver SBRT treatments, in order to perform a detailed investigation of the dosimetric impact of intrafraction motion during actual clinical liver SBRT treatments. Methods and materials Method overview Fig. 1 summarizes the methods of this study. Fluoroscopic kV images acquired during VMAT-based liver SBRT (Fig. 1a) were used to estimate the 3D trajectory of an implanted gold marker [17,18] (Fig. 1b, c), which was in turn used to reconstruct the actually delivered target dose distribution [24] (Fig. 1d). Patients, planning, and treatments This study includes six patients with liver metastases who received SBRT delivered by VMAT in three fractions between March 2011 and February 2013, following our standard SBRT treatment protocol. Guided by ultrasound, each patient had 2–3 cylindrical gold markers (1 mm 3 mm) implanted close to the metastases. The clinical target volume (CTV) was delineated in the mid-ventilation phase of a 4DCT scan with 3 mm slice thickness. The planning
target volume (PTV) was formed by adding 5 mm margins in the left–right (LR) and anterior–posterior (AP) directions and 10 mm in the cranio-caudal (CC) direction. Using the Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA, USA) a 6MV VMAT plan with 5–6 arcs was designed to give a CTV mean dose of 100% and minimum target doses of 95% (CTV) and 67% (PTV), i.e. a non-uniform PTV dose distribution with 50% higher dose in the central part [5–8]. Following a risk-adapted strategy, fraction doses of 12.56 Gy or 16.75 Gy were prescribed to the 67% isodose surface (tightly enclosing the PTV surface) resulting in 100% dose levels of 18.75 Gy or 25 Gy. The arc fields spanned between 134° and 234° and were delivered with a maximum dose rate of 600 MU/min. The degree of VMAT intensity modulation can be characterized by the number of monitor units (MU) per Gy (Table 1) or the mean MLC aperture area relative to the total area covered by the arcs, which ranged from 15% to 47% for the six VMAT plans (average 27%). The table rotation was 15° for four out of six arcs for one patient (Patient 2) and 0° for all other arcs. A Stereotactic Body Frame (Elekta, Crawley, UK) or an in-house modified breast board was used for immobilization. Abdominal compression was used for all patients. Table 1 summarizes relevant patient and image data. The treatments were delivered using five different Trilogy accelerators equipped with an MV PortalVision AS500 or AS1000 portal imager and an On-Board Imager (OBI) (Varian Medical Systems). Prior to treatment, a cone-beam CT scan was used for markerbased patient setup. Continuous portal images (12.8 Hz for AS500, 7.5 Hz for AS1000, 160 cm source-imager-distance (SID)) and orthogonal kV images (5.0 Hz, 125 kV, 80 mA, 11–19 ms, fullfan bow-tie filter, 180 cm SID) were acquired throughout the treatment delivery. The kV field size (Table 1) was chosen as the patient specific minimum size that covered all markers plus a 1.5–2 cm margin as seen from all kV imaging angles of the arc field. The effective dose from intra-treatment kV imaging was coarsely estimated to be 3–6 mSv per fraction from a CBCT scan mode with similar settings (11.0 Hz, 125 kV, 80 mA, 25 ms, full-fan bow-tie filter) by scaling the CBCT dose (9.1 mSv) with the number of images, CC field size, and exposure duration (Table 1). The image and motion data of one patient were previously reported as part of a pilot study on image-based intrafraction motion monitoring (Patient 1, who is identical to Patient 6 in Ref. [14]) while image data for the remaining five patients and all dose reconstruction data and KIM accuracy data are new in this study. Cross-scattering of the MV treatment beam onto the kV imager degraded the quality of the intra-treatment kV images. As a coarse indicator of the VMAT field size and thus the amount of MV irradiation hitting the patient, Table 1 shows the average jaw size for each treatment plan. Since the kV image quality also depended on the kV beam path length through the patient, Table 1 presents the longest kV water equivalent path length (WEPL) that passed through the monitored gold marker. Intra-treatment kV imaging was not attempted for three other liver SBRT patients treated with VMAT in the same period as the
Fig. 1. Method overview.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
3
P.R. Poulsen et al. / Radiotherapy and Oncology xxx (2014) xxx–xxx
Table 1 Patient and image data, including mean CTV dose, number of metastases, effective dose per fraction by intra-treatment kV imaging (E), maximum water equivalent path length of the kV imaging rays through the selected marker (Max kV WEPL), and the kV field size and MV jaw field size averaged over all arc fields of the VMAT plan. Patient
Sex
Age (years)
Metastases
CTV dose (Gy)
VMAT fields
MU/Gy
kV images per field
kV field size (cm x cm)
E (mSv)
Max kV WEPL (cm)
MV jaws (cm cm)
MV imager
1 2 3 4 5 6
F M F M F M
60 79 75 73 68 56
6 4 2 1 1 1
3 18.75 3 18.75 3 25 3 25 3 25 3 25
5 6 5 5 5 5
200 319 167 186 171 161
415 532 439 503 466 434
12.5 3.6 7.5 5.7 15.2 4.2 6.8 2.7 9.2 4.8 8 5.5
3 6 3 3 3 4
28.3 34.8 31.3 37.1 31.8 35.9
17.9 14.8 14.0 14.0 12.1 10.7 8.0 6.9 7.9 5.7 9.6 8.6
AS1000 AS1000 AS1000 AS500 AS1000 AS500
six included patients because a combination of large field sizes and long kV WEPL indicated the possibility of poor intra-treatment kV image quality.
Intra-treatment 3D motion After the treatments, an automatic in-house developed segmentation algorithm was used to determine the position of the gold marker closest to the isocenter in all kV images with a resolution of 1 pixel (0.22 mm) [25]. The auto-segmentation was manually inspected in all kV images and corrected if the segmented position deviated more than 3 pixels (0.7 mm) from the marker center. The marker segmentation provided the projected 2D motion of the selected marker in the rotating kV imager plane (Fig. 1b). After correction for gantry and image system sag [26] the 2D motion was used to estimate the 3D marker motion in the patient (Fig. 1c) by a probability based method [18,27] that uses a fitted 3D Gaussian probability density function to estimate the unknown marker position perpendicular to the kV imager (and thus the full 3D position) for each kV image. The group systematic error M in the intra-treatment marker position relative to the planned position and the standard deviation of the systematic error R and random error r were calculated following Ref. [28]. For some of the non-coplanar arcs of Patient 2, the marker was not visible by eye in the kV images in one or two periods of 1–2 s duration because of excessive MV cross-scatter onto the kV imager. The internal 3D marker motion during these short periods was reconstructed from the motion of an abdominal marker block (RPM, Varian Medical System) by use of a linear motion correlation model. The correlation model was established using the internal 3D marker motion of the segmentable kV images of the arc field and had an estimated accuracy of 1 mm or less. The accuracy of the kV-based 3D marker position estimation was determined from the continuous MV images as follows: For the four patients imaged with an AS1000 portal imager (Table 1), the marker was manually segmented in all MV images with marker visibility at one treatment fraction. The kV-estimated 3D trajectory was projected onto the portal imager and the kV position estimation error was calculated as the difference (scaled to isocenter distance) between the projected positions and the segmented marker positions in the MV imager (Fig. 1, bottom). This validation of the kV position estimation was not performed for the two patients imaged with an AS500 portal imager because the marker contrast was too poor for reliable segmentation. The probability based 3D position estimation method not only provided the target position for each kV image, but also the rootmean-square (rms) error of the target position estimation [27,29] in the unknown direction perpendicular to the kV imager. The rms over all kV images of this rms estimation error was calculated for each patient. Note that this error (unlike the MV determined error above) only contains contributions from the probability based 3D position estimation method and not from segmentation uncertainties or image system sag.
Dose reconstruction For all fractions, the 3D marker motion was used to reconstruct the delivered target dose by an experimentally validated method [24,30] that models the motion of a rigid target as multiple isocenter shifts in an in-house built software program (Matlab) and utilizes the clinical treatment planning system for the dose calculation (Eclipse 11.0, Anisotropic Analytical Algorithm (AAA), Varian Medical Systems). The calculations assumed that the MLC positions as function of gantry angle were equal to the planned MLC positions. A 1.5 mm bin width was used, i.e. all target positions within a 1.5 1.5 1.5 mm3 cube were given the same isocenter shift. For each fraction and treatment course, the reduction of the CTV mean dose (DDmean) and of the minimum dose to 95% of the CTV (DD95) relative to the treatment plan was calculated. Effects of anatomy changes on reconstructed dose The motion including dose reconstruction accounts for rigid target motion, interplay effects, and physical path length changes, but neglects target deformations, radiological path length changes, and target shifts relative to the gold marker [24]. While all dose calculations were performed by applying the isocenter shifts in the midventilation phase of the 4DCT scan, breathing induced anatomy deformation could have been included by distributing the dose calculations among all 4DCT phases [19,20]. However, this would restrict the dose calculations to the target shifts observed in the 4DCT scan, whereas a much wider range of shifts may occur at treatment [31]. The maximum erroneous effect of performing all dose reconstructions in the mid-ventilation phase was estimated as follows for each patient: First, the original treatment plan was recalculated in the full inhale and full exhale phases, which provided reference dose distributions in these extreme phases including the effects of anatomical changes. Next, the plan was recalculated in the mid-ventilation phase, but with the isocenter shifted corresponding to the marker shift between the mid-ventilation phase the extreme phases, which mimicked the dose reconstruction method in these phases. The root-mean-square (rms) dose difference between the dose distributions of the two calculations (in the extreme 4DCT phase and the mid-ventilation phase, respectively) was calculated by in-house developed software as a measure of the dose reconstruction error caused by the most extreme anatomical changes observed in the 4DCT scans. While this analysis quantified the maximum errors in the spatial dose distribution within the patient volume it did not assess errors in the reconstructed CTV dose distribution stemming from deformations of the CTV or shifts of the CTV relative to the monitored marker. Results Intra-treatment 3D motion Fig. 2 shows an example of the 3D intra-treatment liver tumor motion determined by KIM at a treatment fraction with relatively
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
4
Motion and dose during VMAT liver SBRT
regular breathing. At this fraction, the intra-treatment peak-topeak motion range was 5.4 mm (LR), 14.2 mm (CC), and 8.4 mm (AP). A gradual drift during the treatment resulted in relatively large baseline motion ranges (i.e. range of intra-field mean positions) of 2.1 mm (LR), 2.7 mm (CC), and 1.8 mm (AP). The mean position error (i.e. the patient-specific systematic error) was 3.5 mm (LR), 2.5 mm (CC), 0.0 mm (AP), and 4.3 mm (3D). The relatively large mean position error was caused by an intrafraction tumor shift between the setup CBCT imaging and treatment start. Table 2 presents the mean motion ranges and position errors for all patients. The average mean 3D position error was 2.9 mm for single fractions and 2.1 mm for treatment courses. The largest motion ranges during any treatment fraction was 8.7 mm (LR), 35.4 mm (CC), 9.7 mm (AP), and 35.7 (3D). The population based group mean error and the standard deviations of systematic and random errors in the LR, CC, and AP directions were: M = 1.0 mm, 1.1 mm, and 0.7 mm; R = 1.1 mm, 1.7 mm, and 1.0 mm; and r = 1.2 mm, 3.9 mm, and 1.6 mm, respectively. The gold marker was visible within the MLC aperture in 22% of the validation MV images (Table 2). The high rate of marker occlusion by the MLC was partly caused by marker implantation outside the CTV in order to avoid tumor cell spread. The mean [and maximum] rms position estimation error of the KIM method, as determined by the MV images, was 0.36 mm [0.61 mm] parallel to the kV imager and 0.47 mm [0.81 mm] perpendicular to the kV imager (Fig. 2, bottom graphs, Table 2). The probability based
position estimation method itself estimated an rms position estimation error perpendicular to the kV imager of 0.34 mm when averaged over all 18 fractions (Table 2). Dose reconstruction Fig. 3 presents the reconstructed doses and resulting CTV dosevolume-histograms for Patient 1. Compared to the planned dose distribution the delivered doses were blurred due to motion, shifted due to target position errors, and redistributed due to interplay effects. The average reduction in mean CTV dose at treatment was 2.2% (Table 2). The CTV DD95 was in mean 6.0% for single fractions and 5.3% for treatment courses (Table 2), and it correlated with the mean 3D position error (p < 0.001, Pearson, Fig. 4). In some cases, DD95 was however substantial even though the mean position error was modest (Fig. 4, points labeled Patient 1 and Patient 2). The reason for this was large random errors caused either by large mean position errors in different directions at the three treatment fractions (Patient 1 course dose, see Fig. 3) or by large respiratory motion (Patient 2 fraction and course doses, see Table 2). Effects of anatomy changes on reconstructed dose The mean rms difference between the dose distributions reconstructed with isocenter shifts in the mid-ventilation phase and the
Fig. 2. Three upper graphs: Intra-treatment target motion (black) relative to the planned position (red line) in left–right (LR), cranio-caudal (CC), and anterior–posterior (AP) direction as estimated by kilovoltage intrafraction motion monitoring during delivery of five VMAT fields for Patient 5 at fraction 2. Two lower graphs: Projection of the estimated 3D positions onto the MV imager (black) compared with the actual position measured in the MV images (red). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Table 2 Mean over three fractions of the intra-treatment marker motion ranges and magnitude of mean target position errors. Root-mean-square errors (rmse) of kilovoltage intrafraction monitoring (KIM) measured by MV imaging (parallel and perpendicular to the kV imager) and estimated by the KIM method itself (perpendicular to the kV imager). Reduction in CTV D95 and CTV mean dose for each treatment course. Pt
1 2 3 4 5 6 All
Motion range (mm) Total LR/CC/AP
Baseline LR/CC/AP
3.7/12.7/2.4 2.9/27.1/8.2 5.4/8.8/5.2 5.6/21.1/8.7 4.7/12.1/7.7 6.2/15.6/6.5 4.8/16.2/6.4
0.8/1.1/0.4 0.6/2.1/0.8 1.9/1.1/1.0 0.5/1.5/0.7 1.2/1.5/1.0 0.5/1.4/0.7 0.9/1.4/0.8
Magnitude of error (mm)
MV images Total
% with marker
Parallel to kV mean [max]
Perpendicular to kV mean [max]
2958 4586 3143 – 3355 – 14042
23 3 32 – 40 – 22
0.43 0.41 0.30 – 0.30 – 0.36
0.41 0.41 0.38 – 0.67 – 0.47
LR/CC/AP 1.0/3.0/1.3 1.3/1.0/0.7 1.3/0.6/0.4 0.5/3.0/1.5 3.1/3.4/1.8 1.1/0.8/1.1 1.4/2.0/1.1
KIM rmse in MV images (mm)
[0.50] [0.61] [0.35] [0.48] [0.61]
[0.49] [0.60] [0.40] [0.81] [0.81]
KIM rmse (mm)
CTV DD95 (%)
CTV DDmean (%)
0.28 0.28 0.38 0.33 0.42 0.36 0.34
4.0 9.5 0.3 6.0 11.1 1.0 5.3
1.6 5.4 0.8 2.2 3.0 0.2 2.2
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
P.R. Poulsen et al. / Radiotherapy and Oncology xxx (2014) xxx–xxx
5
Fig. 3. Left: Planned and actually delivered dose distributions shown as >95% dose color wash in a coronal plane with the GTV (red contour) and PTV (blue) for Patient 1. Right: Corresponding CTV dose volume histograms. DD95 denotes the reduction in the minimum dose to 95% of the CTV at treatment. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 4. 3D target position error versus reduction in CTV D95 for individual treatment fractions (small black circles) and treatment courses (gray).
reference dose distributions in the extreme 4DCT phases was 0.55% and 0.64% of the prescribed dose when including all voxels with doses larger than 10% dose and 90% dose, respectively. These numbers represent errors in the reconstructed dose distribution caused by large anatomical changes in general and in high dose regions, respectively. The rms dose difference correlated with the 3D target shift |Dr| between the mid-ventilation and extreme phases (p < 0.0001, Pearson) and was to first order equal to 0.10%/mm |Dr| (in general) and 0.16%/mm |Dr| (high dose regions). The 3D tumor motion range was considerably larger during treatment delivery (mean: 17.3 mm; range: 7.3–35.7 mm) than in the planning 4DCT scan (mean: 7.6 mm; range: 4.0–14.7 mm) showing that the 4DCT scans failed to capture the range of anatomy changes occurring at treatment. Linear extrapolation of the rms dose difference in the 4DCT scans to the most extreme target shift observed during treatment (|Dr|max = 27.5 mm, Patient 2, fraction 3) yielded rms dose calculation errors of 2.7% (in general) and 4.4% (high dose regions) for this extreme position and of 0.57% (in general) and 0.92% (high dose regions) when time averaged over all target positions at this particular fraction. Discussion For the first time, kilovoltage intrafraction motion monitoring with dose reconstruction was demonstrated for respiratory moving targets. At 18 liver SBRT treatment fractions, the 3D intra-treatment target motion was accurately estimated throughout VMAT
delivery using the standard IGRT equipment of a linear accelerator. Validation MV imaging demonstrated sub-millimeter localization accuracy for four treatment fractions, while the inherent error estimation of the single-imager localization method indicated similar accuracy for the fractions not monitored by MV imaging. The observed motion was in accordance with previous studies of liver tumor motion [12,14,31,32]. The baseline drift in the example in Fig. 2 agrees with MR observations of systematic intrafraction drifts of the liver [33], but it was not a general motion trend in the current study. Despite CBCT guided patient setup large 3D mean position errors occasionally occurred (up to 6.6 mm) leading to substantial deviations between planned and delivered CTV doses (Fig. 4). Post-treatment review of the online CBCT setup for the treatments with largest position errors showed that, within the uncertainties of marker-based CBCT setup [11], the online marker-based registration was performed correctly. The occasionally large position errors primarily resulted from baseline shifts occurring between IGRT imaging and treatment completion. The correlation between DD95 and the 3D mean position error suggests that a relatively simple correction strategy involving inter-field couch adjustments in case of baseline shifts may give substantial dosimetric improvements. However, this correction strategy would not mitigate dosimetric errors caused by large respiratory motion (e.g. Patient 2 in Fig. 4). Instead, KIM may be combined with real-time marker segmentation and used for real-time motion adaptation during VMAT delivery [34–35]. Despite large CTV D95 reductions, all patients in this study received the prescribed dose since the dose was prescribed to the 67% dose level, whereas the minimum CTV dose at any fraction [course] was 71% [76%]. The large CTV D95 reductions were caused by CTV motion relative to the central high dose region of the PTV. The CTV dose would be less susceptible to intrafraction motion if a uniform PTV dose were prescribed rather than a PTV dose that gradually decreased from 95% to 67% outside the CTV [5–10]. However, in order to be isotoxic a uniform PTV dose would mean a lower prescribed dose to the hypoxia-prone tumor center than with the peaked PTV dose prescription. Because of large uncertainties in current modeling of the tumor control probability and normal tissue complication probability, in particular for non-uniform dose distributions, it is not obvious which prescription strategy that would be optimal. While the presented KIM motion data and dose reconstruction method may be used in combination with biological modeling to investigate dose prescription optimization under consideration of intrafraction tumor motion, this subject is beyond the scope of the current study. A limitation of the KIM method is the additional imaging dose, which was comparable to the dose of a CBCT scan in this study (Table 1). The number of single images acquired per second (five)
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
6
Motion and dose during VMAT liver SBRT
was one order of magnitude lower than with the RTRT system [12], which uses stereoscopic imaging at 30 Hz for internal marker motion monitoring. Real-time tracking of lung tumors has been performed with the Vero4DRT system based on an external-internal motion model that was established prior to treatment by stereoscopic imaging at 6.25 Hz or 12.5 Hz for 40 s [36] and monitored by 1 Hz stereoscopic imaging during 6–7 field deliveries of 32 s duration [37]. It gives up to 1448 single kV images plus additional imaging for 1–2 model updates at each fraction [36]. The current study obtained direct internal motion monitoring with 2100–3200 kV images per fraction (Table 1). The direct internal (surrogate) monitoring allows use of KIM for targets without external motion correlation such as the prostate [17]. The current KIM implementation suffers from excessive MV cross-scatter onto the kV imager, which made the marker invisible in some kV images for one patient, challenged robust automatic marker segmentation in general, and caused us to omit kV imaging for three patients treated with VMAT-based liver SBRT in the same period. Improved kV imager read-out that includes only the kV exposure period (11–19 ms) rather than the entire period between kV images (200 ms) would reduce the duration of MV scatter exposure of the kV images by one order of magnitude. In this study, the reconstructed dose included interplay effects as well as shifts and blurring of the target dose caused by systematic and random intrafraction motion. Separation of the dose effects into these components were beyond the scope of this study. However, we did not observe any noticeable interplay effects such as cold spots within high dose regions. This is in agreement with recent studies of respiration induced interplay effects during VMAT SBRT [19,20,23]. Only negligible interplay effects may be expected as the delivery time at each treatment fraction covered more than 100 breathing cycles. The dose reconstruction in this study accounted for target motion, interplay effects, and physical path length changes, it did not account for breathing induced anatomy deformations since all calculations were performed in the mid-ventilation phase of the planning 4DCT scan. This simplification caused estimated rms errors in the dose distribution in the patient of up to 1% for the patient with largest intra-treatment motion. However, within the reconstructed dose distribution we assumed that the CTV kept the same shape and position relative to the gold marker as in the mid-ventilation planning CT phase. Based on deformable registration between exhale and inhale CT volumes Velec et al. studied dose accumulation over the breathing cycle in liver SBRT and found that, compared to deformable registration, simple rigid liver-to-liver registration resulted in up to 8% discrepancies in the accumulated minimum tumor dose [38]. However, in their rigid registration dose accumulation, the authors assumed that the tumor motion was identical to the liver center-of-mass motion, which was obviously not a valid assumption since deformable registration revealed differences between tumor and liver center-of-mass motion of up to 6 mm (LR), 10 mm (CC), and 7 mm (AP) for these patients [38]. In the current study, we used an implanted gold marker (and not the liver center-of-mass) as a surrogate for the CTV position, which is a much more reasonable assumption [39]. Therefore, the lack of deformation modeling will have a much smaller impact on the accumulated tumor dose than in Ref. [38]. In conclusion, kilovoltage intrafraction motion monitoring with dose reconstruction for VMAT-based liver SBRT was demonstrated for the first time revealing large dosimetric impact of intrafraction motion. Since the motion measurements and dose reconstruction rely on standard equipment and software for IGRT and treatment planning, respectively, the methods have the potential for widespread use. The improved reporting of the actually delivered CTV dose could lead to better understanding of clinical dose–response relationships in radiation oncology.
Conflict of interest This work was partially supported by a research grant from Varian Medical Systems, Inc. (Palo Alto, CA). P.R. Poulsen receives royalties from a patent related to this work. Acknowledgments This work was supported The Danish Cancer Society, CIRRO – The Lundbeck Foundation Center for Interventional Research in Radiation Oncology, and Varian Medical Systems, Inc., Palo Alto, CA. References [1] Yu CX. Intensity-modulated arc therapy with dynamic multileaf collimation: an alternative to tomotherapy. Phys Med Biol 1995;40:1435–49. [2] Otto K. Volumetric modulated arc therapy: IMRT in a single gantry arc. Med Phys 2008;35:310–7. [3] Wang C, Luan S, Tang G, Chen DZ, Earl MA, Yu CX. Arc-modulated radiation therapy (AMRT): a single-arc form of intensity-modulated arc therapy. Phys Med Biol 2008;53:6291–303. [4] Benedict SH, Yenice KM, Followill D, et al. Stereotactic body radiation therapy: the report of AAPM Task Group 101. Med Phys 2010;37:4078–101. [5] Lax I, Blomgren H, Näslund I, Svanström R. Stereotactic radiotherapy of malignancies in the abdomen. Methodological aspects. Acta Oncol 1994;33:677–83. [6] Lax I. Target dose versus extra target dose in stereotactic radiosurgery. Acta Oncol 1993;32:453–7. [7] Lax I, Panettieri V, Wennberg B, et al. Dose distribution in SBRT of lung tumors: comparison between two different treatment planning algorithms and MonteCarlo simulation including breathing motion. Acta Oncol 2006;45:978–88. [8] Hoyer M, Roed H, Traberg Hansen A, et al. Phase II study on stereotactic body radiotherapy of colorectal metastases. Acta Oncol 2006;45:823–30. [9] de Pooter JA, Wunderink W, Mendez Romero A, et al. PTV dose prescription strategies for SBRT of metastatic liver tumours. Radiother Oncol 2007;85:260–6. [10] Molinelli S, de Pooter J, Mendez Romero A, et al. Simultaneous tumour dose escalation and liver sparing in Stereotactic Body Radiation Therapy (SBRT) for liver tumours due to CTV-to-PTV margin reduction. Radiother Oncol 2008;87:432–8. [11] Worm ES, Høyer M, Fledelius W, Nielsen JE, Larsen LP, Poulsen PR. On-line use of three-dimensional marker trajectory estimation from cone-beam computed tomography projections for precise setup in radiotherapy for targets with respiratory motion. Int J Radiat Oncol Biol Phys 2012;83:e145–51. [12] Shirato H, Seppenwoolde Y, Kitamura K, et al. Intrafractional tumor motion: lung and liver. Semin Radiat Oncol 2004;14:10–8. [13] Berbeco RI, Neicu T, Rietzel E, et al. A technique for respiratory-gated radiotherapy treatment verification with an EPID in cine mode. Phys Med Biol 2005;50:3669–79. [14] Worm ES, Høyer M, Fledelius W, Poulsen PR. Three-dimensional, timeresolved, intrafraction motion monitoring throughout stereotactic liver radiation therapy on a conventional linear accelerator. Int J Radiat Oncol Biol Phys 2013;86:190–7. [15] Azcona JD, Li R, Mok E, et al. Development and clinical evaluation of automatic fiducial detection for tumor tracking in cine megavoltage images during volumetric modulated arc therapy. Med Phys 2013;40:031708. [16] Adamson J, Wu Q. Prostate intrafraction motion evaluation using kV fluoroscopy during treatment delivery: a feasibility and accuracy study. Med Phys 2008;35:1793–806. [17] Ng JA, Booth JT, Poulsen PR, et al. Kilovoltage intrafraction monitoring for prostate intensity modulated arc therapy: first clinical results. Int J Radiat Oncol Biol Phys 2012;84:e655–61. [18] Poulsen PR, Cho B, Keall PJ. A method to estimate mean position, motion magnitude, motion correlation, and trajectory of a tumor from cone-beam CT projections for image-guided radiotherapy. Int J Radiat Oncol Biol Phys 2008;72:1587–96. [19] Rao M, Wu J, Cao D, et al. Dosimetric impact of breathing motion in lung stereotactic body radiotherapy treatment using image-modulated radiotherapy and volumetric modulated arc therapy. Int J Radiat Oncol Biol Phys 2012;83:e251–6. [20] Li X, Yang Y, Li T, et al. Dosimetric effect of respiratory motion on volumetricmodulated arc therapy-based lung SBRT treatment delivered by TrueBeam machine with flattening filter-free beam. J Appl Clin Med Phys 2013;14:195–204. [21] Belec J, Clark BG. Monte Carlo calculation of VMAT and helical tomotherapy dose distributions for lung stereotactic treatments with intra-fraction motion. Phys Med Biol 2013;58:2807–21. [22] Feygelman V, Stambaugh C, Zhang G, et al. Motion as a perturbation: measurement-guided dose estimates to moving patient voxels during modulated arc deliveries. Med Phys 2013;40:021708.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007
P.R. Poulsen et al. / Radiotherapy and Oncology xxx (2014) xxx–xxx [23] Stambaugh C, Nelms BE, Dilling T, et al. Experimentally studied dynamic dose interplay does not meaningfully affect target dose in VMAT SBRT lung treatments. Med Phys 2013;40:091710. [24] Poulsen PR, Schmidt ML, Keall P, Worm ES, Fledelius W, Hoffmann L. A method of dose reconstruction for moving targets compatible with dynamic treatments. Med Phys 2012;39:6237–46. [25] Fledelius W, Worm E, Elstrom UV, et al. Robust automatic segmentation of multiple implanted cylindrical gold fiducial markers in cone-beam CT projections. Med Phys 2011;38:6351. [26] Cho B, Poulsen PR, Sloutsky A, Sawant A, Keall P. First demonstration of combined kV/MV image-guided real-time DMLC target tracking. Int J Radiat Oncol Biol Phys 2009;74:859–67. [27] Poulsen PR, Cho B, Keall PJ. Real-time prostate trajectory estimation with a single imager in arc radiotherapy: a simulation study. Phys Med Biol 2009;54:4019–35. [28] van Herk M. Errors and margins in radiotherapy. Semin Radiat Oncol 2004;14:52–64. [29] Poulsen PR, Cho B, Keall PJ. Prediction of position estimation errors for 3D target trajectories estimated from cone-beam CT projections. In: Proceedings of XVIth international conference on the use of computers in radiation therapy, Amsterdam, 31/5/2010–3/6/2010. [30] Ravkilde T, Keall PJ, Grau C, Høyer M, Poulsen PR. Time-resolved dose reconstruction by motion encoding of volumetric modulated arc therapy fields delivered with and without dynamic multi-leaf collimator tracking. Acta Oncol 2013;52:1497–503. [31] Worm ES, Høyer M, Fledelius W, Hansen AT, Poulsen PR. Variations in magnitude and directionality of respiratory target motion throughout full
[32]
[33]
[34]
[35] [36]
[37]
[38]
[39]
7
treatment courses of stereotactic body radiotherapy for tumors in the liver. Acta Oncol 2013;52:1437–44. Gabrys D, Kulik R, Trela K, Slosarek K. Dosimetric comparison of liver tumour radiotherapy in all respiratory phases and in one phase using 4DCT. Radiother Oncol 2011;100:360–4. von Siebenthal M, Székely G, Lomax AJ, Cattin PC. Systematic errors in respiratory gating due to intrafraction deformations of the liver. Med Phys 2007;34:3620–9. Poulsen PR, Fledelius W, Cho B, Keall P. Image-based dynamic multileaf collimator tracking of moving targets during intensity-modulated arc therapy. Int J Radiat Oncol Biol Phys 2012;83:e265–71. Falk M, af Rosenschöld PM, Keall P, et al. Real-time dynamic MLC tracking for inversely optimized arc radiotherapy. Radiother Oncol 2010;94:218–23. Akimoto M, Nakamura M, Mukumoto N, et al. Predictive uncertainty in infrared marker-based dynamic tumor tracking with Vero4DRT. Med Phys 2013;40:091705. Nakamura M, Akimoto M, Mukumoto N, et al. Influence of predictive modelling duration on the predictive accuracy of IR-marker-based dynamic tumour tracking. Radiother Oncol 2013;108:S320. Velec M, Moseley JL, Eccles CL, et al. Effect of breathing motion on radiotherapy dose accumulation in the abdomen using deformable registration. Int J Radiat Oncol Biol Phys 2011;80:265–72. Seppenwoolde Y, Wunderink W, Wunderink-van Veen SR, Storchi P, Mendes Romero A, Heijmen BJ. Treatment precision of image-guided liver SBRT using implanted fiducial markers depends on marker-tumour distance. Phys Med Biol 2011;56:5445–68.
Please cite this article in press as: Poulsen PR et al. Kilovoltage intrafraction motion monitoring and target dose reconstruction for stereotactic volumetric modulated arc therapy of tumors in the liver. Radiother Oncol (2014), http://dx.doi.org/10.1016/j.radonc.2014.05.007