Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: Background and purpose: The treatment quality of pencil beam scanned (PBS) proton therapy to mobile
Received 25 June 2016 tumour treatments can be compromised due to interplay effects. The aim of this work is to systematically
Received in revised form 12 September evaluate the effectiveness of rescanning for liver tumour treatments for a commercial PBS delivery sys-
2016
tem.
Accepted 25 September 2016
Available online 7 October 2016
Materials and methods: Plans were calculated to patient specific ITV’s (2GyRBE), using spot spacings of 4
and 8 mm for 1- and 3-field plans. 4D dose calculations were performed using regular and irregular
motion extracted from nine 4DCT(MRI) liver datasets with 4 different starting phases. Up to 19 times
Keywords:
Scanned proton therapy
adaptive-scaled layered and volumetric rescanning were simulated using beam profiles and delivery
Moving targets dynamics of a commercial proton therapy system.
4D dose calculation Results: For small (10 mm) motions, 3-field plans achieved CTV HI’s (D5–D95) to within 8.5% (80th per-
Interplay effect centile) of the static case without rescanning. For larger motions, volumetric rescanning resulted in 4.5%
Rescanning improved HI in comparison to layered, but requires 5 times longer treatment times and is more sensitive
4DMRI to detailed plan characteristics and delivery dynamics. Increased spot spacings were found to reduce sen-
sitivity to interplay and reduce delivery times by 60%, whilst reduced energy switching times decreased
treatment time by up to 75% for volumetric rescanning without however improving plan quality.
Conclusion: For the investigated proton therapy system, rescanning can help recover dose homogeneity
under conditions of motion but, particularly for motions over 10 mm, should be combined with addi-
tional motion mitigation techniques.
Ó 2016 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 121 (2016) 281–287
Pencil Beam Scanned proton therapy (PBS) is becoming one of be reduced if motion amplitude is controlled, such as by breath-
the most attractive radiotherapy techniques, with promising clini- hold [8] or beam gating [9–12] strategies or, at the other end of
cal outcomes being demonstrated for various indications [1–3]. the spectrum of complexity, by tumour tracking [13–15].
However, its efficacy for the treatment of moving tumours (e.g. Alternatively, rescanning [16,17] can be regarded as a relatively
lung, liver and other tumour sites in the upper abdomen or pelvic straightforward way to deal with the interplay effect, since it is
region) is still a matter of concern, mainly due to the interplay only dependent on statistical averaging and is independent of
effects [4]. In addition to motion induced dose blurring effects at any extra equipment, implementation or active cooperation from
the edge of the target volume, which can be efficiently addressed the patient.
using an internal target volume (ITV) [5], additional over- and Although the effectiveness of rescanning for mobile tumours
under-dosage within the target volume, due to the interplay of has been demonstrated in a number of simulations and experi-
delivery dynamics and anatomical motion, cannot be compensated ments [18–23], its performance is highly machine-specific, simply
by margins only. Therefore, selecting and implementing additional due to its sensitivity to small variations in the relative timelines of
motion mitigation approaches for PBS are essential for assuring the beam delivery and respiratory tumour motion [18]. In addition,
advantages of this technique in terms of delivery quality. beam size, as well as machine specific scanning parameters, spot
Various strategies have been proposed to mitigate motion distance, dose rate, lateral position switching time, energy layer
effects for PBS based treatments [6,7]. For instance, interplay can switching time etc., all have a significant impact on the timeline
of each delivered spot and can therefore result in substantial differ-
ences in the final dose distribution and motion mitigation efficacy.
⇑ Corresponding author at: WBBA/014, Paul Scherrer Institut, 5232 Villigen-PSI,
Switzerland.
Consequently, it is important to study such effects using as accu-
E-mail address: ye.zhang@psi.ch (Y. Zhang). rate a simulation of the actual delivery conditions as possible, so
http://dx.doi.org/10.1016/j.radonc.2016.09.011
0167-8140/Ó 2016 Elsevier Ireland Ltd. All rights reserved.
282 PBS proton rescanning for liver tumour
as to be able to obtain a meaningful quantification of the interplay temporal resolution of 2–3 Hz. The mean, inter-cycle averaged
effects, as well as to investigate the optimal parameters for apply- motion magnitudes for motion A, B and C were of the order of
ing a robust plan. 10, 15 and 20 mm with mean/range periods, calculated using
In this study, we have systematically evaluated the effective- Fourier analysis of the breathing signals, of 3.3(2.9–4.0),
ness of different liver tumour treatments using the delivery charac- 6.3(5.2–7.2), 5.3(4.7–6.3) seconds respectively.
teristics of the Varian ProBeam PBS system, through the use of
dedicated 4D dose calculations (4DDC) which consider the specific 4D dose calculation
parameters of this system (raster scanning with variable dose rate),
together with a comprehensive motion database which not only The beam profiles and scanning parameters used for all 4DDC’s
takes into account realistic patient geometries and deformable are those of the Varian ProBeam PBS system. As such, proton pencil
motion, but also includes inter-cycle breathing variability. In par- beams have been modelled using an energy dependent parameter-
ticular, we have compared the efficacy of different rescanning isation of integral depth dose curves in water and a two-
approaches and treatment plan scenarios, as well as the dosimetric dimensional Gaussian representation of lateral profiles with beams
effects of motion irregularity. sizes (sigma) in air of 4–5.5 mm at isocentre (depending on the
energy). For each field, pencil beam separations orthogonal to the
beam direction of either 4 or 8 mm have been used, together with
Materials and methods
energy switching times (for 3.5 MeV, corresponding to range steps
of 3.9–8.6 mm (75–245 MeV) in water)) of 700, 500, 200 and
Patient and motion data
100 ms. The faster energy switching times have been included in
Nine 4DCT(MRI) datasets [24] have been used in this study, this analysis in order to analyse the potential impact of energy
generated from three 4DCT data sets of three liver patients switching time on interplay effects, rescanning efficiency and
(denoted as I, II and III respectively), each additionally modulated treatment duration. For lateral motion of pencil beams, raster scan-
by three different motion scenarios (denoted as A, B and C) ning has been modelled by including the magnetic scanning time
extracted from a 4DMRI motion library of liver motion using in the delivery timeline calculation. Raster scanning speeds are
dynamic image fusion [24]. In order to produce these datasets, 20 mm/ms and 5 mm/ms for the two directions respectively, with
the mechanical correspondence of liver meshes in MRI and CT were maximum dose rates being set to be around 2 1010 protons/s
firstly determined by defining landmarks which move similarly (energy dependent). For consecutive pencil beams spaced by more
across different livers. Then, the simulated 4DCT(MRI) are created than 10 mm, the beam is switched-off in between. In addition, the
by warping these deformation fields, extracted from 4DMRI, to a duration of each spot irradiation depends on the fluence (number
static 3DCT data set. The tumour locations and characteristics of of protons per spot resulting from the field specific optimisation
each patient are shown in Fig. 1. CTV volumes at End-of- process) and the dose rate (protons/s) of the proton beam at the
Exhalation (EE, the reference phase) were 403, 264 and 122 cc given energy, as the ProBeam system varies dose rate on an energy
for Patients I, II and III respectively. The extracted liver motions layer-by-layer basis, such that the pencil beam with the smallest
(with respect to the reference EE phase of the first breathing cycle) weight in the layer can be delivered with a duration of at least
have also been shown in Fig. 1, as acquired from 4DMRI with a 3 ms.
Fig. 1. Components used for generating 4DCT(MRI) datasets for 4D dose calculations.
Y. Zhang et al. / Radiotherapy and Oncology 121 (2016) 281–287 283
Rescanning Results
Two rescanning strategies have been implemented in this study
The effectiveness of rescanning
– layered and volumetric rescanning (LS and VS). For layered res-
canning, spots within one iso-energy layer are visited several times Due to differences in tumour geometry and location, this 4D
until the planned fluences per pencil beam are completely deliv- analysis has considered a large variation of plan characteristics.
ered, before the energy is changed for the next layer. In contrast, The calculated fields contained between 775 and 2138 individually
for volumetric rescanning, instead of repeating the scanning pat- weighted proton pencil beams, between 14 and 26 energy layers
tern within one energy layer, the whole target volume (all energy (ranging between 77 and 178.5 MeV), and between 4 and
layers) is visited in each rescan. 12 1010 protons-per-field to deliver 2GyRBE to the gITV. Without
For both approaches, the division of beam weight (protons rescanning (Fig. 2), three-field plans achieved HI’s within
delivered for each spot per rescan) is based on a scaling model in 7.2/8.5/12.0% (median/80th percentile/maximum) of the static case
which all spot weights are divided uniformly by a given factor, for all three investigated patients with the 10 mm motion trace
with this scaled beam weight being applied in each rescan. How- (scenario A), independent of motion irregularities. In contrast, for
ever, it should be noted that within any one energy layer, opti- the larger motions (scenarios B and C), HI’s for the three-field plans
mised PBS fields can have a large dynamic range of pencil beam were 15.8/18.1/20.2 worse than the static case when rescanning is
weights [25], and thus when multiple rescans are applied, there not employed. In addition, by taking into account the four different
is an increasing number of low weighted pencil beams which can starting phases of beam delivery, statistically significant differences
be difficult to apply. For this reason, an adaptive-scaling algorithm (P 0.01) between LS and VS were found. For the same motions
has been used, which can compensate this by comparing the scaled and three-field plans, volumetric rescanning (VS) provided mean
beam weight with a pre-defined threshold (3.5 million protons) improvements of 4.5% in HI in comparison to layered rescanning
to assure that all beam weights are above threshold and are deliv- (LS). LS could at best achieve HI to within only 9.2/10.9/13.5%
erable. Pencil beams are not further visited once their defined (median/80th/max) of the static case, whilst VS could reduce fur-
beam weight has been fully achieved in previous rescanning. ther to within 5.1/6.6/10.3% (median/80th/max). However, this
comes at a cost of, on average, 5 times longer treatment times for
VS. For the smaller motion (A), LS and VS were able to achieve com-
parable plan homogeneity. However, for the larger motions (B and
4D treatment planning and plan evaluation
C), increased variation in HI, as a function of patient and increasing
scan number, was found for VS.
In order to differentiate the dose inhomogeneity induced by
interplay effects from those of dose blurring at the field edges, a
geometric Internal Target Volume (gITV) was calculated for each The impact of spot spacing and plan configuration
patient from the 4DCT(MRI) datasets by the Clinical Target Volume When spot spacing was enlarged from 4 mm to 8 mm, treat-
(CTV) from each phase of the first breathing cycle of each motion ment times for a single scan could be reduced by 60%, mainly
scenario (see Fig. 1). PTV was assumed to be equal to CTV without due to the higher dose rates which were achieved as a result of
considering patient setup error. Three single-field-plans (using the the higher average weight per pencil beam for the 8 mm plans.
field directions depicted in Fig. 1) and a three-field-plan were cal- In addition, without rescanning, there was also an around 20%
culated for each of the nine cases, using the reference CT at the EE reduction of the interplay effect when using the larger spot spac-
phase. For all fields and plans, the Single Field Uniform Dose ing. Compared to the three-field-plans, the single-field-plans have
(SFUD) approach was used [26], ensuring a homogenous coverage almost double the magnitude of interplay effect if no motion mit-
of the target from all individual fields and a prescription dose of igation is applied, especially for larger motion, with HI’s as high as
2GyRBE to the case specific gITV was defined for all plans. Layered 80% being calculated. Moreover, the uncertainty bands (20th–80th
and volumetric rescanning with 1–19 times (1 rescan is equal percentile) for these single-field plans were generally larger than
to no rescan) have then been simulated in the 4DDC using the for the three-field-plans, indicating a higher sensitivity of single-
above mentioned parameters of the ProBeam system. The 4DDCs field plans to field direction, patient and motion irregularity. Sig-
have also been performed for both regular (repeated motion of nificant differences (P 0.01) between LS and VS have been found
the first breathing cycle) and irregular motion (based on multiple for all cases, being however less pronounced for the smaller
breathing cycles), and have been simulated each with four differ- motion. The different results indicate the sensitivity of 4D treat-
ent starting phases: end exhalation (0%), middle inhalation (25%), ments on the details of the delivery dynamics.
end inhalation (50%) and middle exhalation (75%).
Dose Volume Histograms (DVHs) were calculated for each 4D
The impact of motion starting phase
plan, from which the homogeneity index (HI) of the dose within
the CTV was measured in terms of the difference between the D5 Our statistical analysis indicates significant differences
and D95 (D5–D95). Since a relatively large gITV has been used (P 0.01) between 4D plans calculated with different starting
which covers the range of tumour motion, any deterioration in phases. However, there was no direct correlation between nor-
the HI value will be mainly due to over- and under-dosages result- malised HI variation and rescan number for either LS or VS. As
ing from residual interplay effects. Furthermore, in order to obtain shown by boxplots in Fig. 3(c) and (d), the dosimetric impact
statistical significance, a two-sided Wilcoxon rank sum test (signif- was generally more pronounced and varied for VS than LS. The
icance level = 5%, paired) has been performed between the result- median value of the normalised variation of up to 30% were
ing HI values from the different rescanning approaches, plan observed for individual subjects and specific rescan number. This
configurations, delivery starting phases, motion patterns and indicated that LS tends to be more predictable, whilst not quite
energy switching times. Additionally, we have used a normalised so effective as VS.
quantification of the variation of HI, defined as the difference
between maximum and minimum HI divided by the mean HI, to
The impact of motion irregularity
quantify the influence of starting phase and energy switching time.
The higher this variation, the higher the impact expected for the As shown in Fig. 4, both LS and VS yielded statistically signifi-
analysed parameter. cant (P 0.01) higher HI values (1% in mean over all scenarios)
284 PBS proton rescanning for liver tumour
Fig. 2. HI of 4D plans (in CTV) using layered (red) or volumetric (blue) rescanning with a function of scan number, motion scenarios and plan configurations. Colour-bands
represent the 20th–80th percentile of HIs from 3 different patient subjects, 4 different starting phases, 2 types of motion pattern, and 3 field directions (for single-field-plan).
Dots in red and blue show data of each 4D plan.
for irregular motion than regular motion, indicating that an under- the energy switching time could have a larger impact on treatment
estimation of motion effects could be obtained if motion is mod- quality on VS than LS. These results are also visible in Fig. 5(b),
elled as a perfectly repeatable pattern (e.g. the 4DCT approach). where HI variations can be seen to be more pronounced for VS
This disparity (shown by uncertainty bands of 20th–80th per- when energy switching time was reduced to 100 ms (red).
centile) can even reach ±5% for individual patients or rescanning
scenarios, without however a clear correlation with rescan num-
ber. This effect seems to be more pronounced for VS than LS and Discussion
maybe due to the longer treatment times which may increase
the probability of encountering more substantial motion We have performed a comprehensive 4D treatment planning
variations. study to quantify interplay effects, and the effectiveness of rescan-
ning, for PBS proton liver treatments using the Varian ProBeam
system. In total, more than 100,000 4D dose calculations have been
Impact of energy switching time
performed, analysing 10 different variables that could affect the 4D
As shown in Fig. 5(c) and (d) for the three-field-plan, by adjust- results, including patient geometry, tumour motion, motion pat-
ing the energy switching time (dE) from 700 ms successively down tern, spot spacing, plan configuration, field direction, starting
to 500, 200 and 100 ms, the median treatment time could be phase, rescan approach, scan number and energy switching time.
reduced by 25%, 63% and 75% respectively for volumetric rescan- In addition, in contrast to the conventional 4DCT approach, by
ning. Delivery time reductions were less pronounced for layered- using 4DCT(MRI) datasets, breathing irregularity has also been
rescanning (<10%). However, no evidence could be found to indi- taken into account, providing a unique opportunity to evaluate
cate systematic improvements in plan homogeneity for either LS the effects of organ motion on PBS proton therapy as a function
or VS (Fig. 5(a) and (b)) when shorter energy switching times were of different 4D delivery parameters and for realistic motion
employed. Similar results have also been found for the two single- scenarios.
field-plan scenarios, as shown in Suppl. 3. On the contrary, Our results show that the effects of the interplay effect, and res-
statistical analysis indicated a significant (P 0.01) increase of canning as a mitigation approach, are both patient specific and
HI values for VS when dE was reduced from 700 ms to 500 or highly dependent on tumour geometry and motion amplitude, as
200 ms, with the median normalised variation of HI values being well as plan and delivery parameters. For the 10 mm motion (sce-
30% for VS, but only around 10% for LS, indicating that reducing nario A) without rescanning, the 80th-percentile of HI’s (HI80) were
Y. Zhang et al. / Radiotherapy and Oncology 121 (2016) 281–287 285
Fig. 3. (a and b) HI in CTV (three-field-plan) and (c and d) its normalized variation (the ratio of range and mean), as a function of scan number, rescan mode and starting
phases. Colour-bands present the 20th–80th percentile of HIs from the 9 cases, each with 2 types of motion pattern (regular and irregular).
Fig. 4. Dosimetric difference of HIs from 4D plans (three-field-plans) with regular and irregular motion pattern. Colour-bands present the 20th–80th percentile of the
differences from the 9 cases, each with 4 different starting phases.
within 8.5% of the HI for the static case for the three-field-plans, robustness to motion. For the larger motions however (scenarios
but were only within 16.5% for single-field-plans. However, for sin- B and C), interplay can only be partially mitigated using rescanning
gle field plans, using rescanning with an appropriate rescan factor or multiple fields alone. For the three-field-plans without rescan-
(>10) could improve HI80 to within 6.1% of the static case for lay- ning, HI80 increased by 18.1% compared to the static case, and by
ered rescanning (LS) and to within 1.9% for volumetric rescanning 31.8% for single-field plans. Even though these can be reduced with
(VS). Combining three-field plans with x10 rescanning can further x10 rescanning, single-field HI80 values were still 14.3% higher
improve HI80 to within 5.2% of the static case for LS, and within than the respective static cases for LS, and 9.1% higher for VS. By
1.8% for VS. These findings indicate the importance of applying res- combining three-field-plans with 10 rescanning however, HI80
canning and/or multiple-field-plans (cf. [27]) to increase plan can get to within 10.9% of the static case for LS and 6.6% for VS.
286 PBS proton rescanning for liver tumour
Fig. 5. (a and b) HIs in CTV, (c and d) treatment time (three-field-plan) and (e and f) the normalized variation (the ratio of range and mean) of HIs with a function of rescan
numbers, rescan mode and energy switching times. Colour-bands present the 20th–80th percentile of 4D plans from the 9 cases, each with 2 types of motion pattern and 4
different starting phases (only for HI).
Moreover, variations in HI (shown by 20th–80th percentile colour difficult to deliver on the ProBeam system, due to the resulting
bands) also increase as a function of patient geometry and starting number of spots that fall below minimally deliverable spot fluence.
phase, indicating the necessity for combining rescanning with As such, this scenario has not been investigated further here.
other motion mitigation strategies (eg. gating) [27] in order to Motion irregularity has also been found to have a noticeable influ-
obtain reliable or predictable mitigation for large motions. Never- ence on all patient scenarios, but is especially pronounced for lar-
theless, for the larger motions, VS on average achieves a 4.5% ger motions (scenarios B and C), and generally has more influence
improved HI in comparison to LS, but requires 5 times longer to on volumetric rescanning than layered rescanning, due to the
deliver the same plan, with its efficacy to mitigate interplay also longer treatment times which increase the possibility of encoun-
being less predictable as a function of rescan number. Besides HI, tering larger motion variations. However, this could also be due
V95 (shown by Suppl. 1) shows a similar tendency, whilst the to the gITV being defined from only a single breathing cycle from
mean dose in the CTV (Suppl. 2) stays within ±2% (without rescan- the original planning 4DCT, which therefore does not include such
ning) and ±1% (with rescanning) of the static reference. This obser- motion variations. More work needs to be performed in this
vation is in good agreement with the results from previous studies direction.
based on the machine parameters of other centres [12,19,20,22]. Moreover, shortening energy switching time would certainly
Using different plan configurations can change the time struc- decrease treatment delivery time (particularly for VS), but there
ture of each spot delivery. Although the spot position and energy is no evidence from this work to show that plan quality would
of the single-field plans are the same as the corresponding field actually be improved. In contrast, more pronounced fluctuations
from the three-field plan, spot fluences are 3 times higher, and in dose homogeneity have been observed for volumetric rescan-
are thus associated with a longer beam-on time. As such, the dosi- ning when energy switching time was reduced to 100 ms. In addi-
metric impacts will be different with respect to motion. Moreover, tion, one should note that we have here considered a conventional
using different spot spacing also ultimately changes the time struc- fractionation scheme (2 Gy/fraction) and the 4DDC’s have been
ture of the delivery, and is therefore expected to have an impact on performed for one fraction only. However, fractionation helps to
the sensitivity of a plan to the interplay effect. For example, a smal- mitigate motion effects, at least if it is assumed that the starting
ler spot spacing requires a larger number of spots but a lower num- phase for delivery of each fraction is randomly distributed over
ber of protons per spot, thus increasing the scanning time by the breathing cycle. This effect though will become progressively
decreasing the effective dose rate for the system studied in this less effective if hypo-fractionation schedules are used.
work. However, with raster scanning, there is only very little differ- Furthermore, we need to mention that our 4D dose calculation
ence in in-layer scanning time as a function of pencil beam spacing, simulated a continuous plan delivery, which means for a given res-
so that higher effective dose rates and shorter treatment times are can plan, there is no waiting time to synchronise each scan and
expected for the plan with larger spot spacing. On the other hand, each breathing cycle. Changing the scan number will completely
when more rescans are applied, the dose rate is nevertheless lim- change the timeline of each spot with respect to the motion pat-
ited by the pre-defined threshold of the spot weight within one tern. Therefore, there is no absolute guarantee that more rescan-
energy layer (3.5 million protons in the relevant energy range), ning will always provide an improved quality. As we have tried
so that differences in treatment time between plans with different to show here, there are many factors affecting the homogeneity
spot spacing become less and less pronounced. Interestingly, in and conformity of the delivered dose under conditions of motion,
this work, we have found that the combination of three field plans and it is not a surprise to us that rescanning cannot mitigate all.
and 4 mm spot spacings under conditions of rescanning may be It is also worth mentioning that there is a limited scan number that
Y. Zhang et al. / Radiotherapy and Oncology 121 (2016) 281–287 287