- Email: [email protected]
Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate
Radiotherapy and Oncology xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Radiotherapy and Oncology journal homepage: www.thegreenjournal.com
Original article
Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate Ghulam Murtaza a,b, Jakob Toftegaard a, Ehsan Ullah Khan b,c, Per Rugaard Poulsen a,⇑ a
Department of Oncology, Aarhus University Hospital, Denmark; b Department of Physics, International Islamic University; and c Muslim Youth University, Islamabad, Pakistan
a r t i c l e
i n f o
Article history: Received 26 July 2016 Received in revised form 31 October 2016 Accepted 3 November 2016 Available online xxxx Keywords: IGRT Prostate motion MLC tracking
a b s t r a c t Purpose: To improve MLC tracking of prostate VMAT plans by dynamic rotation of the collimator to align the MLC leaves with the dominant prostate motion direction. Methods: For 22 prostate cancer patients, two dual arc VMAT plans were made with (1) fixed collimators (45° and 315°) and (2) a rotating collimator that aligned the MLC leaves with the dominant prostate motion direction (population-based first principal component). The fixed and rotating collimator plan quality was compared using selected dose–volume indices. Next, MLC tracking treatments were simulated with 695 patient-measured prostate traces. The MLC exposure error (under- and overexposed MLC area in beam’s eye view) was calculated as a surrogate for the MLC tracking error. Finally, motion including dose reconstruction was performed for 35 motion traces for one patient, and the root-meansquare dose error was compared with the MLC exposure error. Results: Rotating collimator VMAT plans were of similar quality as the fixed collimator plans, but significantly improved MLC tracking with 33% lower MLC exposure errors (p 0.0001). The reductions in MLC exposure error correlated significantly with dose error reductions. Conclusion: Prostate VMAT plans with rotating collimator were of similar quality as fixed collimator plans, but more suitable for MLC tracking with significantly better agreement between planned and delivered dose distributions. MLC tracking for prostate cancer patients can therefore be improved without the requirement of additional efforts or hardware changes. Ó 2016 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology xxx (2016) xxx–xxx
Accurate dose delivery is crucial for radical prostate radiotherapy to ensure optimal target coverage with maximal healthy tissue sparing. While volumetric modulated arc therapy (VMAT) [1] allows efficient delivery of highly conformal dose distributions, organ motion during treatment delivery may compromise the prostate dose [2]. In recent clinical studies, intrafraction motion during prostate VMAT delivery on a conventional linear accelerator was mitigated in real-time by either multileaf collimator (MLC) tracking [3] or gating [4]. The gating involves treatment interruption and couch correction if the prostate motion exceeds a pre-defined action level such as 3 mm excursion for at least 5 s [4]. Gating and MLC tracking both improve the consistency between the planned and delivered doses [5,6], but tracking has the advantages of not prolonging the treatment time [3], not requiring radiotherapist interaction, and not having an action level below which motion is not accounted for. ⇑ Corresponding author at: Aarhus University Hospital, Nr Brogade 44, 8000 Aarhus C, Denmark. E-mail address: [email protected] (P.R. Poulsen).
A recent comparison of MLC and couch tracking for prostate and lung VMAT treatments showed that the largest residual dose errors with tracking occurred for MLC tracking of prostate motion [7]. The reason was that MLC tracking cannot compensate for small persistent prostate shifts perpendicular to the MLC leaves [7]. Improved MLC tracking would be expected if the MLC leaves at each gantry control point were aligned with the major component of prostate motion in beam’s eye view, thereby minimizing target motion perpendicular to the MLC leaves. This would require knowledge of the dominant prostate motion direction at the time of treatment planning and the ability to align the collimator angle to this direction as a function of the gantry angle during treatment delivery. Several studies have shown that intrafraction prostate motion over a patient population has a preferred motion direction with a tendency of simultaneous cranial-anterior motion or caudalposterior motion [8–12]. Hence, a population-based prostate motion directionality may be used as an approximate motion direction for the individual patient in the population. Furthermore, VMAT treatments with dynamic collimator rotations can be delivered in the non-clinical Developer Mode of a TrueBeam accelerator (Varian Medical Systems,Palo Alto,CA). The use of dynamic
http://dx.doi.org/10.1016/j.radonc.2016.11.004 0167-8140/Ó 2016 Elsevier Ireland Ltd. All rights reserved.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
2
Improved prostate MLC tracking with rotating collimator VMAT
collimator rotation during VMAT has previously been investigated for improved dose conformity in paraspinal stereotactic body radiotherapy [13] and online adaptation to interfraction prostate rotations [14]. In this study, we first introduce prostate VMAT planning with a dynamic collimator rotation that aligns the MLC leaves to the population-averaged major prostate motion direction, and then investigate MLC tracking improvements of the rotating collimator VMAT plans in large-scale MLC tracking simulations.
Methods and materials This study consists of three parts as illustrated in Supplementary Fig. 1 and described in the following. First, rotating collimator VMAT plans were produced for a number of patients and the plan objectives were compared with standard fixed collimator VMAT plans (Part A). Next, realistic large-scale MLC tracking simulations were performed with both plan types, and the MLC tracking performance was compared in terms of residual MLC exposure errors (Part B) and dosimetric errors (Part C). Patient selection, treatment planning, and plan comparison In Part A, a retrospective treatment planning study was carried out in order to compare VMAT prostate plans with fixed and rotating collimator. The study included 22 consecutive prostate cancer patients treated in supine position with dual arc VMAT in 39 fractions at Aarhus University Hospital in 2015. As part of the clinical routine, the prostate (clinical target volume for 78 Gy, CTV78), bladder, and rectum were delineated in a planning CT scan. The seminal vesicles (CTV55) were delineated as a target for simultaneous integrated irradiation for nine of the patients. Planning target volumes (PTV78 and PTV55) were created by extending the CTVs by 7 mm in the left–right (LR) and anterior-posterior (AP) directions and by 9 mm in the cranio-caudal (CC) direction. For each patient, two 15MV dual arc VMAT plans with 260° counter-clockwise (CCW) and clockwise (CW) gantry rotation between 130° and 230° were made for a TrueBeam accelerator with a Millennium MLC with 5 mm leaf width (Varian). The first VMAT plan had fixed collimator angles of 45° (CCW) and 315° (CW), which is the standard planning technique in our department. The second VMAT plan had a gantry-dependent dynamically rotating collimator that was adapted to the population-based prostate motion direction as described in the following. The MLC tracking simulations (see next section) were performed with 695 prostate motion traces recorded at 10 Hz with implanted electromagnetic transponders (Calypso, Varian) for 17 patients in supine position at MD Anderson Cancer Center Orlando [15]. All available traces of at least 5 min duration were used. In the rotating collimator VMAT plans, the gantry-dependent collimator angle was selected such that the MLC leaves were aligned with the major component of intrafraction prostate motion for the population. The first 5 min of all 695 traces were concatenated to one long trajectory, for which the first principal component (PC1) was determined by principal component analysis: PC1 = (PCLR,PCCC, PCAP) = (0.001,0.603,0.798) in the LR, CC, and AP directions. Neglecting the small LR motion, PC1 is lying in the sagittal plane with an angle of arctan(PCAP/PCCC) = 53° relative to the cranial direction. Projected to beam’s eye view, PC1 has a constant component PCCC in the cranial direction while its perpendicular component depends on the gantry angle hgantry as PCAPsin(hgantry). Therefore the following gantry-dependent collimator angle hcol was used in the rotating collimator VMAT plans for alignment of the MLC leaves with PC1 in beam’s eye view:
hcol ¼ 90 þ arctanððPCAP sinðhgantry ÞÞ=PCCC Þ ¼ 90 þ arctanð1:323 sinðhgantry ÞÞ The principal component analysis showed that 82.2% of the prostate motion (variance) occurred along PC1, i.e. along the MLC leaves for the rotating collimator plans, while the remaining 17.8% motion occurred either perpendicular to the MLC leaves or in-depth along the beam axis. VMAT planning was performed using the Photon Optimizer (PO) version 136.23.01 in a non-clinical research version of Eclipse 13.6 (Varian) that allowed VMAT planning with collimator rotation between the control points. Table 1 (column 3) summarizes the clinical VMAT planning objectives. The same dose–volume objectives were used for optimization of both VMAT plans for a given patient to allow rigorous comparison between the plans. In Table 1, VnGy and Vx denote the percentage of a volume that receives a dose of at least nGy and x% of the prescribed dose, respectively. Dose–volume histograms (DVHs) of targets and risk organs were used to compare the fixed and rotating collimator VMAT plans by the dose–volume indices in Table 1. The homogeneity index (HI) for PTV78 was calculated as (D2 D98)/Dmean, where Dx is the dose delivered to at least x% of PTV78 and Dmean is the PTV78 mean dose. Statistically significant differences between fixed and rotating collimator VMAT plans were investigated with a two-sided student’s t-test assuming unequal variances and a significance threshold of p = 0.05. Treatment simulations with and without MLC tracking In Part B of the study, VMAT treatment delivery was simulated with and without MLC tracking for both VMAT plan types for all 22 patients and all 695 prostate traces. Mimicking our clinical practice, the simulations assumed image-guided patient setup with prostate alignment to two orthogonal X-ray images. The appropriate treatment start time relative to the image-guided alignment was determined as the recorded mean duration for ten of the patients between setup imaging and treatment start time averaged over seven fractions. VMAT treatment simulations with and without MLC tracking were performed with in–house built simulator software [16] that accurately reproduced the accelerator behavior in a series of experiments with electromagnetic guided MLC tracking on a TrueBeam accelerator [7]. For each simulated treatment, the MLC exposure error was quantified as the MLC shielded area in beam’s eye view that should ideally have been exposed (underexposed area, Au) and the exposed area that should ideally have been shielded (overexposed area, Ao) [17]. The time averaged value of Au + Ao was calculated for each simulated treatment as a surrogate for the motion induced dosimetric errors. Previous studies have found good correlation between Au + Ao and the motion induced dose errors [17,18]. A two-sided student’s t-test assuming unequal variances was used to compare the mean Au + Ao of VMAT plans with fixed and rotating collimator with a significance level of p = 0.05. Dose reconstruction The underlying assumption that the MLC exposure error (mean Au + Ao) is a good surrogate for motion induced dose errors was investigated in Part C of the study by motion including dose reconstruction [19] for a selection of the simulated treatments for the first patient. Dose reconstructions were performed for both tracking and non-tracking treatments with fixed and rotating collimator for 35 of the 695 simulated prostate traces, selected to uniformly cover the full range of MLC exposure errors in the simulations. The dose reconstruction modeled the target motion as multiple isocenter shifts and, for the MLC tracking treatments, by replacing
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
3
G. Murtaza et al. / Radiotherapy and Oncology xxx (2016) xxx–xxx Table 1 Clinical objectives and average values (ranges) of dose–volume indices and MUs as well as p-values comparing fixed and rotating collimator VMAT plans. Structure
Index
Clinical objectives
Fixed collimator VMAT
Rotating collimator VMAT
p-Value
CTV78 PTV78
V95% V90% Dmax (%) HI V95% V70Gy (%) V74Gy (cm3) V70Gy (%) V75Gy (%) MUs
P99.5% P99.0% 6107%
100.0 (99.7–100.0)% 99.6 (98.9–100.0)% 105.4 (104.6–106.6)% 0.11 (0.10–0.13) 99.9 (99.8 – 100.0)% 8.7 (4.5 – 14.0)% 0.30 (0.01 –0.91) cm3 21.8 (8.4–39.3)% 16.2 (6.1–28.0)% 447 (393–492)
100.9 (99.8–100.0)% 99.5 (98.9–100.0)% 105.5 (104.5–106.8)% 0.12 (0.10–0.13) 99.4 (98.4–100.0)% 8.5 (4.8–12.9)% 0.50 (0.05–1.05) cm3 19.3 (7.8–35.4)% 13.4 (4.7–24.4)% 422 (362–495)
0.44 0.32 0.52 0.16 0.02 0.82 0.02 0.36 0.15 0.009
PTV55 Rectum Bladder –
P99.5% 620% 61 cm3 635% 625%
HI = Homogeneity index, MUs = Monitor units.
the planned MLC positions with the MLC positions of the tracking simulations [19]. This treatment plan manipulation was performed by an in-house built computer program (Matlab), while the subsequent dose calculation was done in Eclipse after import of the manipulated plans. The motion-including dose distributions were compared with the static planned dose distributions using two metrics. First, the root-mean-square (RMS) dose error relative to the planned static dose was calculated with inclusion of all dose points with doses above 50% and 90% of the prescription dose. These dose errors represent the motion induced distortion of the planned dose distribution in medium high dose regions (>50% points) and high-dose regions (>90% points), respectively. The RMS dose error was calculated in Matlab using exported dose matrices from Eclipse. Second, the motion induced changes, relative to the planned dose, of D95, D98, D99, V95, V98, and V99 were calculated for CTV78. A possible linear relationship between motion induced dose changes and MLC exposure errors was tested by Pearson’s product-moment correlation coefficient using a significance level of 5%. Results Treatment planning The evaluated DVH parameters for CTV78 and PTV78 were not significantly different between fixed and rotating collimator VMAT plans (Table 1). PTV55 V95% was significantly lower and rectum V74Gy was significantly higher for the rotating collimator plans than the fixed collimator plans (Table 1). These differences were merely a result of using identical objectives when optimizing the two plans. They do not reflect inherent limitation of the rotating collimator VMAT planning, since slight modifications of the
dose–volume objectives resulted in rotating collimator plans of similar quality as the fixed collimator plans. The other rectum and bladder DVH parameters were not significantly different between the two plan types, but the rotating collimator plans had significantly fewer MUs than the fixed collimator plans (Table 1). Treatment simulations The mean duration between setup imaging and treatment start was 81 s and the mean time between the two VMAT fields was 7 s. These times were used for all simulations. The delivery duration for each field was 43.5 s in the actual treatments and in all simulations both with and without MLC tracking. Each simulation thus used a prostate trace length of 175 s (81 s + 43.5 s + 7 s + 43.5 s). As shown in Fig. 1a, MLC tracking greatly reduced the mean MLC exposure error (Au + Ao) for both fixed and rotating collimator VMAT plans. Collimator rotation reduced the mean MLC exposure error relative to fixed collimator VMAT by 33% for MLC tracking treatments (2.23 cm2 versus 3.33 cm2, p 0.0001) and by 12% for non-tracking treatments (4.63 cm2 versus 5.24 cm2, p 0.0001). Collimator rotation reduced the mean MLC exposure error for most, but not all, MLC tracking and non-tracking treatments as seen in Fig. 1b and c. Dose reconstruction Fig. 2 compares the planned dose distribution for Patient 1 with the motion-including doses in simulated treatments with and without MLC tracking for one of the 695 motion traces. For this trace, intrafraction motion led to RMS dose errors of more than 5% in the high dose region (Fig. 2b). MLC tracking recovered the
Fig. 1. (a) Cumulative distribution of MLC exposure errors (mean Au + Ao) in 15290 treatment simulations with fixed (red) and rotating (black) collimator VMAT plans delivered with (solid) and without (dashed) MLC tracking. (b-c) Distribution of the reduction in mean MLC exposure error by rotating collimator VMAT relative to fixed collimator VMAT in simulated deliveries (b) with and (c) without MLC tracking.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
4
Improved prostate MLC tracking with rotating collimator VMAT
dose distribution to a high extent for the fixed collimator VMAT plan and almost completely for the rotating collimator plan (Fig. 2c). Fig. 2d indicates that the MLC exposure error was a good surrogate for the dose error for this plan. The improvement in MLC exposure error with MLC tracking by rotating collimator VMAT was similar or better than the example in Fig 2c (i.e. reduction in mean Au + Ao P 2.2 cm2) in 19.8% of all 15290 simulated treatments, whereas a similar deterioration by the rotating collimator (i.e. increase in mean Au + Ao P 2.2 cm2) occurred in 0.26% of the simulated treatments (see Fig. 1b). The MLC exposure error had highly significant correlation (p 0.0001) with the RMS dose errors for all 140 (35x4) reconstructed doses (Fig. 3) and with the motion induced change of all CTV78 DVH indices except V95 for MLC tracking (Supplementary Fig. 2a–f). Furthermore, the reduction in MLC exposure errors during MLC tracking of rotating collimator VMAT plans relative to fixed collimator plans (summarized in Fig. 1b for all simulations)
was accompanied by highly correlated reductions in the RMS dose errors as seen in Fig. 4a and b. The MLC exposure error reduction of rotating collimator VMAT plans correlated weakly, but significantly with reduced motion induced changes for V98 (Fig. 4c), D98 (Fig. 4d), and D99 (not shown) (Pearson r: 0.34–0.47,p < 0.05), but not for V99 (r = 0.47, p = 0.13), V95 (r = 0.13, p = 0.47), and D95 (r = 0.31, p = 0.07). Discussion This study first demonstrated that dynamic alignment of the collimator angle to the major prostate motion direction during VMAT can reach the same plan objectives as clinical standard VMAT plans with a fixed collimator angle (Table 1). Next, largescale simulations showed that rotating collimator VMAT plans significantly reduced the mean MLC exposure errors of MLC tracking by 33% as compared to fixed collimator plans (Fig. 1). Finally, the
Fig. 2. (a-c) Dose distributions in a CT slice as planned and delivered in simulations with and without MLC tracking for both fixed and rotating collimator VMAT plans. (d) Root-mean-square dose error (for doses > 90%) versus MLC exposure error (mean Au + Ao) for the four treatments in (b-c).
Fig. 3. Root-mean-square dose error in points receiving more than (a) 50% and (b) 90% of the prescription dose versus the MLC exposure error (mean Au + Ao) for 35 prostate traces for Patient 1. The Pearson correlation coefficient r is indicated. All correlations had p 0.001.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
G. Murtaza et al. / Radiotherapy and Oncology xxx (2016) xxx–xxx
5
Fig. 4. Relation between reduction in MLC exposure error (mean Au + Ao) during MLC tracking by changing from fixed to rotating collimator VMAT and reduction in rootmean-square dose errors (a and b) or in the absolute difference between planned and delivered (c) V98 and (d) D98 for CTV78. Pearson correlation coefficients r and p-values are indicated.
reduction in MLC exposure errors was shown to correlate significantly with improved MLC tracking, i.e. with improved consistency between planned and delivered doses (Figs. 3 and 4). The proportionality between MLC exposure errors and dose errors suggests that the 33% reduction in MLC exposure errors translates into a similar reduction in the dosimetric error of MLC tracking. These findings are important since large residual dose errors of VMAT MLC tracking are particularly prominent for prostate due to persistent shifts perpendicular to the MLC leaves that the MLC aperture cannot adapt to [7]. It leads to systematic errors, which are dosimetrically more costly than the more random errors that typically occur for lung tumor motion [7]. A solution to this problem is offered by the rotating collimator VMAT approach suggested in this study, which exploits that prostate motion has a generic motion directionality across a population [8–12]. Several alternative methods to improve VMAT MLC tracking of target motion perpendicular to the MLC leaves have been proposed. One possibility is to reduce the plan complexity [20–21], but it has been shown to be far less effective for prostate MLC tracking [20] than for lung tumor tracking [21]. Another strategy is to use tracking Y-jaws instead of the MLC leaf sides to define the field aperture borders, thereby circumventing the discrete
aperture adaptation to motion perpendicular to the MLC leaves [22]. While this may eliminate the MLC exposure errors for strictly rectangular fields it is less powerful for more modulated fields, where the extreme field borders (which can be formed by the Yjaws instead of leaf sides) typically only constitute a small fraction of the total field border formed by MLC leaf sides. For the VMAT plans in the current study, the Y-jaws can in mean replace 12.1% (fixed collimator) and 10.6% (rotating collimator) of the total field border formed by MLC leaf sides. Dynamic tracking of the Y-jaws would thus give 10–12% reduction of the MLC exposure error, while rotating collimator VMAT planning yielded a mean reduction of 33%. Rotating collimator VMAT planning and Y-jaw tracking are not excluding each other, but could be used simultaneously with added value. As seen in Fig. 1 the rotating collimator VMAT plans also reduced the MLC exposure error in non-tracking treatments. The reason for this is that the field border formed by the MLC leaf sides in general was considerably longer than the field border formed by the MLC leaf ends (on average 26.4 cm versus 15.9 cm for fixed collimator VMAT and 27.2 cm versus 16.3 cm for rotating collimator VMAT). It means that a prostate shift perpendicular to the MLC leaves during non-tracking treatments in general gave rise to lar-
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
6
Improved prostate MLC tracking with rotating collimator VMAT
Fig. 5. Reduction in mean MLC exposure error (Au + Ao) during MLC tracking of rotating collimator VMAT plans relative to fixed collimator plans for each of the 695 prostate motion traces for Patient 1 (red), for all other patients (blue), and averaged over all patients (black). The prostate traces were sorted from lowest to highest MLC exposure error reduction for Patient 1. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
ger MLC exposure errors than a prostate shift of the same magnitude parallel to the MLC leaves. Since prostate motion mainly occurred parallel to the MLC leaves for the rotating collimator VMAT plans, these plans had smaller MLC exposure errors than the fixed collimator plans even without MLC tracking. The MLC exposure error (Au + Ao) was found to be a good surrogate for dose errors (Figs. 3 and 4a and b). While this was only shown for Patient 1 due to the heavy workload of dose reconstructions, this patient was representative for the population by having MLC exposure error reductions by rotating collimator VMAT that were very close to the population mean for each prostate trace as shown in Fig. 5. All patients had similar trends as Patient 1 in the MLC exposure error reduction as function of prostate trace number (Fig. 5). This is expected since the MLC exposure error reduction is closely linked to the magnitude and directionality of each individual prostate trace. It is reasonable to assume that the MLC exposure errors reductions in general for all patients are accompanied by similar dosimetric improvements as for Patient 1, which is in accordance with previous demonstrations of high correlation between MLC exposure errors and dose errors [17,18]. While the MLC exposure error reductions for MLC tracking of rotating collimator VMAT plans relative to fixed collimator plans correlated well with reduced dose errors (Fig. 4a and b) it only correlated weakly with improvements of some of the CTV dose–volume indices (V98, D98, D99, Fig. 4cand d). This is partly due to the relatively large CTV–PTV margins of 7–9 mm designed for nontracking treatments. With MLC tracking, smaller margins may be applied, which would make the high agreement between the planned and delivered dose distribution of rotating collimator VMAT plans crucial. The importance of accurate dose delivery could also increase with hypofractionation and sculpted inhomogeneous dose distributions as in dose painting and urethra sparing prostate treatments. To assess the effect of margin reduction from the available dose reconstruction data, pseudo-CTVs were created by expanding the current CTV78 by 1–5 mm in 1 mm steps and removing regions that overlapped with the rectum. D95, D98, D99, V95, V98, and V99 were calculated for all pseudo-CTVs for the 140 dose reconstructions of Patient 1 in order to estimate the effect of CTV-to-PTV margins that were 1–5 mm smaller than the applied clinical margins. Supplementary Fig. 2 shows the motion induced changes in the DVH indices as function of the MLC exposure error for the original CTV78 and the 5 mm expansion. Except for V95 for tracking treatments, which only had motion induced mean changes of 0.32%-point or less, all DVH index changes correlated significantly with the MLC exposure error for the original CTV78 and for all pseudo-CTVs. Not surprisingly, the pseudo-CTV expan-
sion had largest impact on the non-tracking treatments, where it resulted in an increasingly better correlation between MLC exposure errors and DVH index changes (Supplementary Fig. 2) and a significant increase in the motion induced mean change of all six DVH indices (Supplementary Fig. 3). It illustrates the need for relatively large margins without tracking to ensure prostate dose coverage. For MLC tracking treatments, the pseudo-CTV expansion did not increase the motion induced DVH index changes (Supplementary Fig. 3) or their correlation with MLC exposure errors. While this illustrates the potential of MLC tracking to maintain the target dose even with reduced margins, the limited number of dose reconstructions and the use of pseudo-CTVs instead of PTV expansions in this study may not allow quantitative assessment of this potential. Although rotating collimator VMAT treatments are not yet clinically available, the plans can be calculated in current research versions of a commercial treatment planning system and delivered in non-clinical mode of modern linear accelerators (TrueBeam Developer Mode). Dynamic alignment of the VMAT collimator angle to the dominant tumor motion direction could be extended from prostate to targets with respiratory motion, where the tumor specific motion directionality at treatment may be estimated from the motion in a planning 4-dimensional CT scan [23]. As a limitation of this study, the isocenter-shift dose reconstruction method neglected rotations and deformations of the prostate and differential motion of healthy tissue relative to the prostate [19]. Despite this the reconstructed dose in the high dose region, including the prostate itself, is likely to be accurate since the included effects of prostate translations and MLC interplay effects typically have much larger dosimetric impact than the neglected effects of prostate rotations and deformations and radiological path length changes of the entrance beam. Conclusion Rotating collimator VMAT plans are more robust to MLC tracking with better agreement between the planned and the delivered dose distribution. MLC tracking for prostate cancer patients can therefore be improved without the requirement of additional efforts or hardware changes. Conflict of interest Aarhus University Hospital receives financial support from Varian Medical Systems.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
G. Murtaza et al. / Radiotherapy and Oncology xxx (2016) xxx–xxx
Acknowledgements This work was supported The Danish Cancer Society and Varian Medical Systems, Inc., Palo Alto, CA. We gratefully thank Wayne Keranen, Varian Medical Systems, for the Eclipse research version with rotating collimator VMAT planning and Katja Langen, Maryland Proton Treatment Center, for the prostate motion database. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.radonc.2016.11. 004. References [1] Otto K. Volumetric modulated arc therapy: IMRT in a single gantry arc. Med Phys 2008;35:310–7. [2] Langen KM, Jones DT. Organ motion and its management. Int J Radiat Oncol Biol Phys 2001;50:265–78. [3] Colvill E, Booth JT, O’Brien RT, Eade TN, Kneebone AB, Poulsen PR, et al. Multileaf collimator tracking improves dose delivery for prostate cancer radiotherapy: results of the first clinical trial. Int J Radiat Oncol Biol Phys 2015;92:1141–7. [4] Keall PJ, Ng JA, O’Brien R, Colvill E, Huang C-Y, Poulsen PR, et al. The first clinical treatment with kilovoltage intrafraction monitoring (KIM): a real-time image guidance method. Med Phys 2015;42:354–8. [5] Colvill E, Poulsen PR, Booth JT, O’Brien R, Ng JA, Keall PJ. DMLC tracking and gating can improve dose coverage for prostate VMAT. Med Phys 2014;41. 091705-1-10. [6] Colvill E, Booth J, Nill S, Fast M, Bedford J, Oelfke U, et al. A dosimetric comparison of real-time adaptive radiotherapy: a multi-institutional study encompassing robotic, gimbaled, multileaf collimator and couch tracking. Radiother Oncol 2016;119:165. [7] Hansen R, Ravkilde T, Worm ES, Toftegaard J, Grau C, Macek K, et al. Electromagnetic guided couch and multileaf collimator tracking on a TrueBeam accelerator. Med Phys 2016;43:2387–98. [8] Poulsen PR, Muren LP, Høyer M. Residual set-up errors and margins in on-line image-guided prostate localization in radiotherapy. Radiother Oncol 2007;85:201–6.
7
[9] Poulsen PR, Cho B, Langen K, Kupelian P, Keall PJ. Three-dimensional prostate position estimation with a single x-ray imager utilizing the spatial probability density. Phys Med Biol 2008;53:4331–53. [10] Tehrani JN, O’Brien RT, Poulsen PR, Keall PJ. Real-time estimation of prostate tumor rotation and translation with a kV imaging system based on an iterative closest point algorithm. Phys Med Biol 2013;58:8517–33. [11] Lin Y, Liu T, Yang W, Yang X, Khan MK. The non-Gaussian nature of prostate motion based on real-time intrafraction tracking. Int J Radiat Oncol Biol Phys 2013;87:363–9. [12] Mutanga TF, de Boer HCJ, Rajan V, Dirkx MLP, Incrocci L, Heijmen BJM. Day-today reproducibility of prostate intrafraction motion assessed by multiple kV and MV imaging of implanted markers during treatment. Int J Radiat Oncol Biol Phys 2012;83:400–7. [13] Zhang P, Happersett L, Yang Y, Yamada Y, Mageras G, Hunt M. Optimization of collimator trajectory in volumetric modulated arc therapy: development and evaluation for paraspinal SBRT. Int J Radiat Oncol Biol Phys 2010;77:591–9. [14] de Boer J, Wolf AL, Szeto YZ, van Herk M, Sonke J-J. Dynamic collimator angle adjustments during volumetric modulated arc therapy to account for prostate rotations. Int J Radiat Oncol Biol Phys 2015;91:1009–16. [15] Langen KM, Willoughby TR, Meeks SL, Santhanam A, Cunningham A, Levine L, et al. Observations on real-time prostate gland motion using electromagnetic tracking. Int J Radiat Oncol Biol Phys 2008;71:1084–90. [16] Toftegaard J, Hansen R, Ravkilde T, Macek K, Poulsen PR. An experimentally validated couch and MLC tracking simulator used to investigate hybrid couchMLC tracking. Med Phys (accepted for publication). [17] Poulsen PR, Fledelius W, Cho B, Keall PJ. Image-based dynamic MLC tracking of moving targets during intensity modulated arc therapy. Int J Rad Onc Biol Phys 2012;83:e265–71. [18] Ravkilde T, Keall PJ, Grau C, Høyer M, Poulsen PR. Time-resolved dose distributions to moving targets during volumetric modulated arc therapy with and without dynamic MLC tracking. Med Phys 2013;40. 111723-1-8. [19] Poulsen PR, Schmidt ML, Keall P, Worm ES, Fledelius W, Hoffman L. A method of dose reconstruction for moving targets compatible with dynamic treatments. Med Phys 2012;39:6237–46. [20] Pommer T, Falk M, Poulsen PR, Keall PJ, O’Brien RT, Rosenschöld PMunck af. The impact of leaf width and plan complexity on DMLC tracking of prostate intensity modulated arc therapy. Med Phys 2013;40. 111717-1-6. [21] Falk M, Larsson T, Keall P, Cho BC, Aznar M, Korreman S, et al. The dosimetric impact of inversely optimized arc radiotherapy plan modulation for real-time dynamic MLC tracking delivery. Med Phys 2012;39:1588–94. [22] Fast MF, Nill S, Bedford JL, Oelfke U. Dynamic tumor tracking using the Elekta Agility MLC. Med Phys 2014;41. 111719-1-5. [23] Worm ES, Høyer M, Fledelius W, Hansen AT, Poulsen PR. Variations in magnitude and directionality of respiratory target motion throughout full treatment courses of stereotactic body radiotherapy for tumors in the liver. Acta Oncol 2013;52:1437–44.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
Contents lists available at ScienceDirect
Radiotherapy and Oncology journal homepage: www.thegreenjournal.com
Original article
Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate Ghulam Murtaza a,b, Jakob Toftegaard a, Ehsan Ullah Khan b,c, Per Rugaard Poulsen a,⇑ a
Department of Oncology, Aarhus University Hospital, Denmark; b Department of Physics, International Islamic University; and c Muslim Youth University, Islamabad, Pakistan
a r t i c l e
i n f o
Article history: Received 26 July 2016 Received in revised form 31 October 2016 Accepted 3 November 2016 Available online xxxx Keywords: IGRT Prostate motion MLC tracking
a b s t r a c t Purpose: To improve MLC tracking of prostate VMAT plans by dynamic rotation of the collimator to align the MLC leaves with the dominant prostate motion direction. Methods: For 22 prostate cancer patients, two dual arc VMAT plans were made with (1) fixed collimators (45° and 315°) and (2) a rotating collimator that aligned the MLC leaves with the dominant prostate motion direction (population-based first principal component). The fixed and rotating collimator plan quality was compared using selected dose–volume indices. Next, MLC tracking treatments were simulated with 695 patient-measured prostate traces. The MLC exposure error (under- and overexposed MLC area in beam’s eye view) was calculated as a surrogate for the MLC tracking error. Finally, motion including dose reconstruction was performed for 35 motion traces for one patient, and the root-meansquare dose error was compared with the MLC exposure error. Results: Rotating collimator VMAT plans were of similar quality as the fixed collimator plans, but significantly improved MLC tracking with 33% lower MLC exposure errors (p 0.0001). The reductions in MLC exposure error correlated significantly with dose error reductions. Conclusion: Prostate VMAT plans with rotating collimator were of similar quality as fixed collimator plans, but more suitable for MLC tracking with significantly better agreement between planned and delivered dose distributions. MLC tracking for prostate cancer patients can therefore be improved without the requirement of additional efforts or hardware changes. Ó 2016 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology xxx (2016) xxx–xxx
Accurate dose delivery is crucial for radical prostate radiotherapy to ensure optimal target coverage with maximal healthy tissue sparing. While volumetric modulated arc therapy (VMAT) [1] allows efficient delivery of highly conformal dose distributions, organ motion during treatment delivery may compromise the prostate dose [2]. In recent clinical studies, intrafraction motion during prostate VMAT delivery on a conventional linear accelerator was mitigated in real-time by either multileaf collimator (MLC) tracking [3] or gating [4]. The gating involves treatment interruption and couch correction if the prostate motion exceeds a pre-defined action level such as 3 mm excursion for at least 5 s [4]. Gating and MLC tracking both improve the consistency between the planned and delivered doses [5,6], but tracking has the advantages of not prolonging the treatment time [3], not requiring radiotherapist interaction, and not having an action level below which motion is not accounted for. ⇑ Corresponding author at: Aarhus University Hospital, Nr Brogade 44, 8000 Aarhus C, Denmark. E-mail address: [email protected] (P.R. Poulsen).
A recent comparison of MLC and couch tracking for prostate and lung VMAT treatments showed that the largest residual dose errors with tracking occurred for MLC tracking of prostate motion [7]. The reason was that MLC tracking cannot compensate for small persistent prostate shifts perpendicular to the MLC leaves [7]. Improved MLC tracking would be expected if the MLC leaves at each gantry control point were aligned with the major component of prostate motion in beam’s eye view, thereby minimizing target motion perpendicular to the MLC leaves. This would require knowledge of the dominant prostate motion direction at the time of treatment planning and the ability to align the collimator angle to this direction as a function of the gantry angle during treatment delivery. Several studies have shown that intrafraction prostate motion over a patient population has a preferred motion direction with a tendency of simultaneous cranial-anterior motion or caudalposterior motion [8–12]. Hence, a population-based prostate motion directionality may be used as an approximate motion direction for the individual patient in the population. Furthermore, VMAT treatments with dynamic collimator rotations can be delivered in the non-clinical Developer Mode of a TrueBeam accelerator (Varian Medical Systems,Palo Alto,CA). The use of dynamic
http://dx.doi.org/10.1016/j.radonc.2016.11.004 0167-8140/Ó 2016 Elsevier Ireland Ltd. All rights reserved.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
2
Improved prostate MLC tracking with rotating collimator VMAT
collimator rotation during VMAT has previously been investigated for improved dose conformity in paraspinal stereotactic body radiotherapy [13] and online adaptation to interfraction prostate rotations [14]. In this study, we first introduce prostate VMAT planning with a dynamic collimator rotation that aligns the MLC leaves to the population-averaged major prostate motion direction, and then investigate MLC tracking improvements of the rotating collimator VMAT plans in large-scale MLC tracking simulations.
Methods and materials This study consists of three parts as illustrated in Supplementary Fig. 1 and described in the following. First, rotating collimator VMAT plans were produced for a number of patients and the plan objectives were compared with standard fixed collimator VMAT plans (Part A). Next, realistic large-scale MLC tracking simulations were performed with both plan types, and the MLC tracking performance was compared in terms of residual MLC exposure errors (Part B) and dosimetric errors (Part C). Patient selection, treatment planning, and plan comparison In Part A, a retrospective treatment planning study was carried out in order to compare VMAT prostate plans with fixed and rotating collimator. The study included 22 consecutive prostate cancer patients treated in supine position with dual arc VMAT in 39 fractions at Aarhus University Hospital in 2015. As part of the clinical routine, the prostate (clinical target volume for 78 Gy, CTV78), bladder, and rectum were delineated in a planning CT scan. The seminal vesicles (CTV55) were delineated as a target for simultaneous integrated irradiation for nine of the patients. Planning target volumes (PTV78 and PTV55) were created by extending the CTVs by 7 mm in the left–right (LR) and anterior-posterior (AP) directions and by 9 mm in the cranio-caudal (CC) direction. For each patient, two 15MV dual arc VMAT plans with 260° counter-clockwise (CCW) and clockwise (CW) gantry rotation between 130° and 230° were made for a TrueBeam accelerator with a Millennium MLC with 5 mm leaf width (Varian). The first VMAT plan had fixed collimator angles of 45° (CCW) and 315° (CW), which is the standard planning technique in our department. The second VMAT plan had a gantry-dependent dynamically rotating collimator that was adapted to the population-based prostate motion direction as described in the following. The MLC tracking simulations (see next section) were performed with 695 prostate motion traces recorded at 10 Hz with implanted electromagnetic transponders (Calypso, Varian) for 17 patients in supine position at MD Anderson Cancer Center Orlando [15]. All available traces of at least 5 min duration were used. In the rotating collimator VMAT plans, the gantry-dependent collimator angle was selected such that the MLC leaves were aligned with the major component of intrafraction prostate motion for the population. The first 5 min of all 695 traces were concatenated to one long trajectory, for which the first principal component (PC1) was determined by principal component analysis: PC1 = (PCLR,PCCC, PCAP) = (0.001,0.603,0.798) in the LR, CC, and AP directions. Neglecting the small LR motion, PC1 is lying in the sagittal plane with an angle of arctan(PCAP/PCCC) = 53° relative to the cranial direction. Projected to beam’s eye view, PC1 has a constant component PCCC in the cranial direction while its perpendicular component depends on the gantry angle hgantry as PCAPsin(hgantry). Therefore the following gantry-dependent collimator angle hcol was used in the rotating collimator VMAT plans for alignment of the MLC leaves with PC1 in beam’s eye view:
hcol ¼ 90 þ arctanððPCAP sinðhgantry ÞÞ=PCCC Þ ¼ 90 þ arctanð1:323 sinðhgantry ÞÞ The principal component analysis showed that 82.2% of the prostate motion (variance) occurred along PC1, i.e. along the MLC leaves for the rotating collimator plans, while the remaining 17.8% motion occurred either perpendicular to the MLC leaves or in-depth along the beam axis. VMAT planning was performed using the Photon Optimizer (PO) version 136.23.01 in a non-clinical research version of Eclipse 13.6 (Varian) that allowed VMAT planning with collimator rotation between the control points. Table 1 (column 3) summarizes the clinical VMAT planning objectives. The same dose–volume objectives were used for optimization of both VMAT plans for a given patient to allow rigorous comparison between the plans. In Table 1, VnGy and Vx denote the percentage of a volume that receives a dose of at least nGy and x% of the prescribed dose, respectively. Dose–volume histograms (DVHs) of targets and risk organs were used to compare the fixed and rotating collimator VMAT plans by the dose–volume indices in Table 1. The homogeneity index (HI) for PTV78 was calculated as (D2 D98)/Dmean, where Dx is the dose delivered to at least x% of PTV78 and Dmean is the PTV78 mean dose. Statistically significant differences between fixed and rotating collimator VMAT plans were investigated with a two-sided student’s t-test assuming unequal variances and a significance threshold of p = 0.05. Treatment simulations with and without MLC tracking In Part B of the study, VMAT treatment delivery was simulated with and without MLC tracking for both VMAT plan types for all 22 patients and all 695 prostate traces. Mimicking our clinical practice, the simulations assumed image-guided patient setup with prostate alignment to two orthogonal X-ray images. The appropriate treatment start time relative to the image-guided alignment was determined as the recorded mean duration for ten of the patients between setup imaging and treatment start time averaged over seven fractions. VMAT treatment simulations with and without MLC tracking were performed with in–house built simulator software [16] that accurately reproduced the accelerator behavior in a series of experiments with electromagnetic guided MLC tracking on a TrueBeam accelerator [7]. For each simulated treatment, the MLC exposure error was quantified as the MLC shielded area in beam’s eye view that should ideally have been exposed (underexposed area, Au) and the exposed area that should ideally have been shielded (overexposed area, Ao) [17]. The time averaged value of Au + Ao was calculated for each simulated treatment as a surrogate for the motion induced dosimetric errors. Previous studies have found good correlation between Au + Ao and the motion induced dose errors [17,18]. A two-sided student’s t-test assuming unequal variances was used to compare the mean Au + Ao of VMAT plans with fixed and rotating collimator with a significance level of p = 0.05. Dose reconstruction The underlying assumption that the MLC exposure error (mean Au + Ao) is a good surrogate for motion induced dose errors was investigated in Part C of the study by motion including dose reconstruction [19] for a selection of the simulated treatments for the first patient. Dose reconstructions were performed for both tracking and non-tracking treatments with fixed and rotating collimator for 35 of the 695 simulated prostate traces, selected to uniformly cover the full range of MLC exposure errors in the simulations. The dose reconstruction modeled the target motion as multiple isocenter shifts and, for the MLC tracking treatments, by replacing
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
3
G. Murtaza et al. / Radiotherapy and Oncology xxx (2016) xxx–xxx Table 1 Clinical objectives and average values (ranges) of dose–volume indices and MUs as well as p-values comparing fixed and rotating collimator VMAT plans. Structure
Index
Clinical objectives
Fixed collimator VMAT
Rotating collimator VMAT
p-Value
CTV78 PTV78
V95% V90% Dmax (%) HI V95% V70Gy (%) V74Gy (cm3) V70Gy (%) V75Gy (%) MUs
P99.5% P99.0% 6107%
100.0 (99.7–100.0)% 99.6 (98.9–100.0)% 105.4 (104.6–106.6)% 0.11 (0.10–0.13) 99.9 (99.8 – 100.0)% 8.7 (4.5 – 14.0)% 0.30 (0.01 –0.91) cm3 21.8 (8.4–39.3)% 16.2 (6.1–28.0)% 447 (393–492)
100.9 (99.8–100.0)% 99.5 (98.9–100.0)% 105.5 (104.5–106.8)% 0.12 (0.10–0.13) 99.4 (98.4–100.0)% 8.5 (4.8–12.9)% 0.50 (0.05–1.05) cm3 19.3 (7.8–35.4)% 13.4 (4.7–24.4)% 422 (362–495)
0.44 0.32 0.52 0.16 0.02 0.82 0.02 0.36 0.15 0.009
PTV55 Rectum Bladder –
P99.5% 620% 61 cm3 635% 625%
HI = Homogeneity index, MUs = Monitor units.
the planned MLC positions with the MLC positions of the tracking simulations [19]. This treatment plan manipulation was performed by an in-house built computer program (Matlab), while the subsequent dose calculation was done in Eclipse after import of the manipulated plans. The motion-including dose distributions were compared with the static planned dose distributions using two metrics. First, the root-mean-square (RMS) dose error relative to the planned static dose was calculated with inclusion of all dose points with doses above 50% and 90% of the prescription dose. These dose errors represent the motion induced distortion of the planned dose distribution in medium high dose regions (>50% points) and high-dose regions (>90% points), respectively. The RMS dose error was calculated in Matlab using exported dose matrices from Eclipse. Second, the motion induced changes, relative to the planned dose, of D95, D98, D99, V95, V98, and V99 were calculated for CTV78. A possible linear relationship between motion induced dose changes and MLC exposure errors was tested by Pearson’s product-moment correlation coefficient using a significance level of 5%. Results Treatment planning The evaluated DVH parameters for CTV78 and PTV78 were not significantly different between fixed and rotating collimator VMAT plans (Table 1). PTV55 V95% was significantly lower and rectum V74Gy was significantly higher for the rotating collimator plans than the fixed collimator plans (Table 1). These differences were merely a result of using identical objectives when optimizing the two plans. They do not reflect inherent limitation of the rotating collimator VMAT planning, since slight modifications of the
dose–volume objectives resulted in rotating collimator plans of similar quality as the fixed collimator plans. The other rectum and bladder DVH parameters were not significantly different between the two plan types, but the rotating collimator plans had significantly fewer MUs than the fixed collimator plans (Table 1). Treatment simulations The mean duration between setup imaging and treatment start was 81 s and the mean time between the two VMAT fields was 7 s. These times were used for all simulations. The delivery duration for each field was 43.5 s in the actual treatments and in all simulations both with and without MLC tracking. Each simulation thus used a prostate trace length of 175 s (81 s + 43.5 s + 7 s + 43.5 s). As shown in Fig. 1a, MLC tracking greatly reduced the mean MLC exposure error (Au + Ao) for both fixed and rotating collimator VMAT plans. Collimator rotation reduced the mean MLC exposure error relative to fixed collimator VMAT by 33% for MLC tracking treatments (2.23 cm2 versus 3.33 cm2, p 0.0001) and by 12% for non-tracking treatments (4.63 cm2 versus 5.24 cm2, p 0.0001). Collimator rotation reduced the mean MLC exposure error for most, but not all, MLC tracking and non-tracking treatments as seen in Fig. 1b and c. Dose reconstruction Fig. 2 compares the planned dose distribution for Patient 1 with the motion-including doses in simulated treatments with and without MLC tracking for one of the 695 motion traces. For this trace, intrafraction motion led to RMS dose errors of more than 5% in the high dose region (Fig. 2b). MLC tracking recovered the
Fig. 1. (a) Cumulative distribution of MLC exposure errors (mean Au + Ao) in 15290 treatment simulations with fixed (red) and rotating (black) collimator VMAT plans delivered with (solid) and without (dashed) MLC tracking. (b-c) Distribution of the reduction in mean MLC exposure error by rotating collimator VMAT relative to fixed collimator VMAT in simulated deliveries (b) with and (c) without MLC tracking.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
4
Improved prostate MLC tracking with rotating collimator VMAT
dose distribution to a high extent for the fixed collimator VMAT plan and almost completely for the rotating collimator plan (Fig. 2c). Fig. 2d indicates that the MLC exposure error was a good surrogate for the dose error for this plan. The improvement in MLC exposure error with MLC tracking by rotating collimator VMAT was similar or better than the example in Fig 2c (i.e. reduction in mean Au + Ao P 2.2 cm2) in 19.8% of all 15290 simulated treatments, whereas a similar deterioration by the rotating collimator (i.e. increase in mean Au + Ao P 2.2 cm2) occurred in 0.26% of the simulated treatments (see Fig. 1b). The MLC exposure error had highly significant correlation (p 0.0001) with the RMS dose errors for all 140 (35x4) reconstructed doses (Fig. 3) and with the motion induced change of all CTV78 DVH indices except V95 for MLC tracking (Supplementary Fig. 2a–f). Furthermore, the reduction in MLC exposure errors during MLC tracking of rotating collimator VMAT plans relative to fixed collimator plans (summarized in Fig. 1b for all simulations)
was accompanied by highly correlated reductions in the RMS dose errors as seen in Fig. 4a and b. The MLC exposure error reduction of rotating collimator VMAT plans correlated weakly, but significantly with reduced motion induced changes for V98 (Fig. 4c), D98 (Fig. 4d), and D99 (not shown) (Pearson r: 0.34–0.47,p < 0.05), but not for V99 (r = 0.47, p = 0.13), V95 (r = 0.13, p = 0.47), and D95 (r = 0.31, p = 0.07). Discussion This study first demonstrated that dynamic alignment of the collimator angle to the major prostate motion direction during VMAT can reach the same plan objectives as clinical standard VMAT plans with a fixed collimator angle (Table 1). Next, largescale simulations showed that rotating collimator VMAT plans significantly reduced the mean MLC exposure errors of MLC tracking by 33% as compared to fixed collimator plans (Fig. 1). Finally, the
Fig. 2. (a-c) Dose distributions in a CT slice as planned and delivered in simulations with and without MLC tracking for both fixed and rotating collimator VMAT plans. (d) Root-mean-square dose error (for doses > 90%) versus MLC exposure error (mean Au + Ao) for the four treatments in (b-c).
Fig. 3. Root-mean-square dose error in points receiving more than (a) 50% and (b) 90% of the prescription dose versus the MLC exposure error (mean Au + Ao) for 35 prostate traces for Patient 1. The Pearson correlation coefficient r is indicated. All correlations had p 0.001.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
G. Murtaza et al. / Radiotherapy and Oncology xxx (2016) xxx–xxx
5
Fig. 4. Relation between reduction in MLC exposure error (mean Au + Ao) during MLC tracking by changing from fixed to rotating collimator VMAT and reduction in rootmean-square dose errors (a and b) or in the absolute difference between planned and delivered (c) V98 and (d) D98 for CTV78. Pearson correlation coefficients r and p-values are indicated.
reduction in MLC exposure errors was shown to correlate significantly with improved MLC tracking, i.e. with improved consistency between planned and delivered doses (Figs. 3 and 4). The proportionality between MLC exposure errors and dose errors suggests that the 33% reduction in MLC exposure errors translates into a similar reduction in the dosimetric error of MLC tracking. These findings are important since large residual dose errors of VMAT MLC tracking are particularly prominent for prostate due to persistent shifts perpendicular to the MLC leaves that the MLC aperture cannot adapt to [7]. It leads to systematic errors, which are dosimetrically more costly than the more random errors that typically occur for lung tumor motion [7]. A solution to this problem is offered by the rotating collimator VMAT approach suggested in this study, which exploits that prostate motion has a generic motion directionality across a population [8–12]. Several alternative methods to improve VMAT MLC tracking of target motion perpendicular to the MLC leaves have been proposed. One possibility is to reduce the plan complexity [20–21], but it has been shown to be far less effective for prostate MLC tracking [20] than for lung tumor tracking [21]. Another strategy is to use tracking Y-jaws instead of the MLC leaf sides to define the field aperture borders, thereby circumventing the discrete
aperture adaptation to motion perpendicular to the MLC leaves [22]. While this may eliminate the MLC exposure errors for strictly rectangular fields it is less powerful for more modulated fields, where the extreme field borders (which can be formed by the Yjaws instead of leaf sides) typically only constitute a small fraction of the total field border formed by MLC leaf sides. For the VMAT plans in the current study, the Y-jaws can in mean replace 12.1% (fixed collimator) and 10.6% (rotating collimator) of the total field border formed by MLC leaf sides. Dynamic tracking of the Y-jaws would thus give 10–12% reduction of the MLC exposure error, while rotating collimator VMAT planning yielded a mean reduction of 33%. Rotating collimator VMAT planning and Y-jaw tracking are not excluding each other, but could be used simultaneously with added value. As seen in Fig. 1 the rotating collimator VMAT plans also reduced the MLC exposure error in non-tracking treatments. The reason for this is that the field border formed by the MLC leaf sides in general was considerably longer than the field border formed by the MLC leaf ends (on average 26.4 cm versus 15.9 cm for fixed collimator VMAT and 27.2 cm versus 16.3 cm for rotating collimator VMAT). It means that a prostate shift perpendicular to the MLC leaves during non-tracking treatments in general gave rise to lar-
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
6
Improved prostate MLC tracking with rotating collimator VMAT
Fig. 5. Reduction in mean MLC exposure error (Au + Ao) during MLC tracking of rotating collimator VMAT plans relative to fixed collimator plans for each of the 695 prostate motion traces for Patient 1 (red), for all other patients (blue), and averaged over all patients (black). The prostate traces were sorted from lowest to highest MLC exposure error reduction for Patient 1. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
ger MLC exposure errors than a prostate shift of the same magnitude parallel to the MLC leaves. Since prostate motion mainly occurred parallel to the MLC leaves for the rotating collimator VMAT plans, these plans had smaller MLC exposure errors than the fixed collimator plans even without MLC tracking. The MLC exposure error (Au + Ao) was found to be a good surrogate for dose errors (Figs. 3 and 4a and b). While this was only shown for Patient 1 due to the heavy workload of dose reconstructions, this patient was representative for the population by having MLC exposure error reductions by rotating collimator VMAT that were very close to the population mean for each prostate trace as shown in Fig. 5. All patients had similar trends as Patient 1 in the MLC exposure error reduction as function of prostate trace number (Fig. 5). This is expected since the MLC exposure error reduction is closely linked to the magnitude and directionality of each individual prostate trace. It is reasonable to assume that the MLC exposure errors reductions in general for all patients are accompanied by similar dosimetric improvements as for Patient 1, which is in accordance with previous demonstrations of high correlation between MLC exposure errors and dose errors [17,18]. While the MLC exposure error reductions for MLC tracking of rotating collimator VMAT plans relative to fixed collimator plans correlated well with reduced dose errors (Fig. 4a and b) it only correlated weakly with improvements of some of the CTV dose–volume indices (V98, D98, D99, Fig. 4cand d). This is partly due to the relatively large CTV–PTV margins of 7–9 mm designed for nontracking treatments. With MLC tracking, smaller margins may be applied, which would make the high agreement between the planned and delivered dose distribution of rotating collimator VMAT plans crucial. The importance of accurate dose delivery could also increase with hypofractionation and sculpted inhomogeneous dose distributions as in dose painting and urethra sparing prostate treatments. To assess the effect of margin reduction from the available dose reconstruction data, pseudo-CTVs were created by expanding the current CTV78 by 1–5 mm in 1 mm steps and removing regions that overlapped with the rectum. D95, D98, D99, V95, V98, and V99 were calculated for all pseudo-CTVs for the 140 dose reconstructions of Patient 1 in order to estimate the effect of CTV-to-PTV margins that were 1–5 mm smaller than the applied clinical margins. Supplementary Fig. 2 shows the motion induced changes in the DVH indices as function of the MLC exposure error for the original CTV78 and the 5 mm expansion. Except for V95 for tracking treatments, which only had motion induced mean changes of 0.32%-point or less, all DVH index changes correlated significantly with the MLC exposure error for the original CTV78 and for all pseudo-CTVs. Not surprisingly, the pseudo-CTV expan-
sion had largest impact on the non-tracking treatments, where it resulted in an increasingly better correlation between MLC exposure errors and DVH index changes (Supplementary Fig. 2) and a significant increase in the motion induced mean change of all six DVH indices (Supplementary Fig. 3). It illustrates the need for relatively large margins without tracking to ensure prostate dose coverage. For MLC tracking treatments, the pseudo-CTV expansion did not increase the motion induced DVH index changes (Supplementary Fig. 3) or their correlation with MLC exposure errors. While this illustrates the potential of MLC tracking to maintain the target dose even with reduced margins, the limited number of dose reconstructions and the use of pseudo-CTVs instead of PTV expansions in this study may not allow quantitative assessment of this potential. Although rotating collimator VMAT treatments are not yet clinically available, the plans can be calculated in current research versions of a commercial treatment planning system and delivered in non-clinical mode of modern linear accelerators (TrueBeam Developer Mode). Dynamic alignment of the VMAT collimator angle to the dominant tumor motion direction could be extended from prostate to targets with respiratory motion, where the tumor specific motion directionality at treatment may be estimated from the motion in a planning 4-dimensional CT scan [23]. As a limitation of this study, the isocenter-shift dose reconstruction method neglected rotations and deformations of the prostate and differential motion of healthy tissue relative to the prostate [19]. Despite this the reconstructed dose in the high dose region, including the prostate itself, is likely to be accurate since the included effects of prostate translations and MLC interplay effects typically have much larger dosimetric impact than the neglected effects of prostate rotations and deformations and radiological path length changes of the entrance beam. Conclusion Rotating collimator VMAT plans are more robust to MLC tracking with better agreement between the planned and the delivered dose distribution. MLC tracking for prostate cancer patients can therefore be improved without the requirement of additional efforts or hardware changes. Conflict of interest Aarhus University Hospital receives financial support from Varian Medical Systems.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004
G. Murtaza et al. / Radiotherapy and Oncology xxx (2016) xxx–xxx
Acknowledgements This work was supported The Danish Cancer Society and Varian Medical Systems, Inc., Palo Alto, CA. We gratefully thank Wayne Keranen, Varian Medical Systems, for the Eclipse research version with rotating collimator VMAT planning and Katja Langen, Maryland Proton Treatment Center, for the prostate motion database. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.radonc.2016.11. 004. References [1] Otto K. Volumetric modulated arc therapy: IMRT in a single gantry arc. Med Phys 2008;35:310–7. [2] Langen KM, Jones DT. Organ motion and its management. Int J Radiat Oncol Biol Phys 2001;50:265–78. [3] Colvill E, Booth JT, O’Brien RT, Eade TN, Kneebone AB, Poulsen PR, et al. Multileaf collimator tracking improves dose delivery for prostate cancer radiotherapy: results of the first clinical trial. Int J Radiat Oncol Biol Phys 2015;92:1141–7. [4] Keall PJ, Ng JA, O’Brien R, Colvill E, Huang C-Y, Poulsen PR, et al. The first clinical treatment with kilovoltage intrafraction monitoring (KIM): a real-time image guidance method. Med Phys 2015;42:354–8. [5] Colvill E, Poulsen PR, Booth JT, O’Brien R, Ng JA, Keall PJ. DMLC tracking and gating can improve dose coverage for prostate VMAT. Med Phys 2014;41. 091705-1-10. [6] Colvill E, Booth J, Nill S, Fast M, Bedford J, Oelfke U, et al. A dosimetric comparison of real-time adaptive radiotherapy: a multi-institutional study encompassing robotic, gimbaled, multileaf collimator and couch tracking. Radiother Oncol 2016;119:165. [7] Hansen R, Ravkilde T, Worm ES, Toftegaard J, Grau C, Macek K, et al. Electromagnetic guided couch and multileaf collimator tracking on a TrueBeam accelerator. Med Phys 2016;43:2387–98. [8] Poulsen PR, Muren LP, Høyer M. Residual set-up errors and margins in on-line image-guided prostate localization in radiotherapy. Radiother Oncol 2007;85:201–6.
7
[9] Poulsen PR, Cho B, Langen K, Kupelian P, Keall PJ. Three-dimensional prostate position estimation with a single x-ray imager utilizing the spatial probability density. Phys Med Biol 2008;53:4331–53. [10] Tehrani JN, O’Brien RT, Poulsen PR, Keall PJ. Real-time estimation of prostate tumor rotation and translation with a kV imaging system based on an iterative closest point algorithm. Phys Med Biol 2013;58:8517–33. [11] Lin Y, Liu T, Yang W, Yang X, Khan MK. The non-Gaussian nature of prostate motion based on real-time intrafraction tracking. Int J Radiat Oncol Biol Phys 2013;87:363–9. [12] Mutanga TF, de Boer HCJ, Rajan V, Dirkx MLP, Incrocci L, Heijmen BJM. Day-today reproducibility of prostate intrafraction motion assessed by multiple kV and MV imaging of implanted markers during treatment. Int J Radiat Oncol Biol Phys 2012;83:400–7. [13] Zhang P, Happersett L, Yang Y, Yamada Y, Mageras G, Hunt M. Optimization of collimator trajectory in volumetric modulated arc therapy: development and evaluation for paraspinal SBRT. Int J Radiat Oncol Biol Phys 2010;77:591–9. [14] de Boer J, Wolf AL, Szeto YZ, van Herk M, Sonke J-J. Dynamic collimator angle adjustments during volumetric modulated arc therapy to account for prostate rotations. Int J Radiat Oncol Biol Phys 2015;91:1009–16. [15] Langen KM, Willoughby TR, Meeks SL, Santhanam A, Cunningham A, Levine L, et al. Observations on real-time prostate gland motion using electromagnetic tracking. Int J Radiat Oncol Biol Phys 2008;71:1084–90. [16] Toftegaard J, Hansen R, Ravkilde T, Macek K, Poulsen PR. An experimentally validated couch and MLC tracking simulator used to investigate hybrid couchMLC tracking. Med Phys (accepted for publication). [17] Poulsen PR, Fledelius W, Cho B, Keall PJ. Image-based dynamic MLC tracking of moving targets during intensity modulated arc therapy. Int J Rad Onc Biol Phys 2012;83:e265–71. [18] Ravkilde T, Keall PJ, Grau C, Høyer M, Poulsen PR. Time-resolved dose distributions to moving targets during volumetric modulated arc therapy with and without dynamic MLC tracking. Med Phys 2013;40. 111723-1-8. [19] Poulsen PR, Schmidt ML, Keall P, Worm ES, Fledelius W, Hoffman L. A method of dose reconstruction for moving targets compatible with dynamic treatments. Med Phys 2012;39:6237–46. [20] Pommer T, Falk M, Poulsen PR, Keall PJ, O’Brien RT, Rosenschöld PMunck af. The impact of leaf width and plan complexity on DMLC tracking of prostate intensity modulated arc therapy. Med Phys 2013;40. 111717-1-6. [21] Falk M, Larsson T, Keall P, Cho BC, Aznar M, Korreman S, et al. The dosimetric impact of inversely optimized arc radiotherapy plan modulation for real-time dynamic MLC tracking delivery. Med Phys 2012;39:1588–94. [22] Fast MF, Nill S, Bedford JL, Oelfke U. Dynamic tumor tracking using the Elekta Agility MLC. Med Phys 2014;41. 111719-1-5. [23] Worm ES, Høyer M, Fledelius W, Hansen AT, Poulsen PR. Variations in magnitude and directionality of respiratory target motion throughout full treatment courses of stereotactic body radiotherapy for tumors in the liver. Acta Oncol 2013;52:1437–44.
Please cite this article in press as: Murtaza G et al. Volumetric modulated arc therapy with dynamic collimator rotation for improved multileaf collimator tracking of the prostate. Radiother Oncol (2016), http://dx.doi.org/10.1016/j.radonc.2016.11.004