Experimentally studied dynamic dose interplay does not meaningfully affect targetdose in VMAT SBRT lung treatmentsCassandra Stambaugh, Benjamin E. Nelms, Thomas Dilling, Craig Stevens, Kujtim Latifi, Geoffrey Zhang,

Eduardo Moros, and Vladimir Feygelman Citation: Medical Physics 40, 091710 (2013); doi: 10.1118/1.4818255 View online: http://dx.doi.org/10.1118/1.4818255 View Table of Contents: http://scitation.aip.org/content/aapm/journal/medphys/40/9?ver=pdfcov Published by the American Association of Physicists in Medicine

Experimentally studied dynamic dose interplay does not meaningfullyaffect target dose in VMAT SBRT lung treatments

Cassandra StambaughDepartment of Physics, University of South Florida, Tampa, Florida 33612

Benjamin E. NelmsCanis Lupus LLC, Merrimac, Wisconsin 53561

Thomas Dilling, Craig Stevens, Kujtim Latifi, Geoffrey Zhang, Eduardo Moros,and Vladimir Feygelmana)

Department of Radiation Oncology, Moffitt Cancer Center, Tampa, Florida 33612

(Received 29 March 2013; revised 20 June 2013; accepted for publication 30 July 2013;published 16 August 2013)

Purpose: The effects of respiratory motion on the tumor dose can be divided into the gradient andinterplay effects. While the interplay effect is likely to average out over a large number of fractions,it may play a role in hypofractionated [stereotactic body radiation therapy (SBRT)] treatments. Thissubject has been extensively studied for intensity modulated radiation therapy but less so for volumet-ric modulated arc therapy (VMAT), particularly in application to hypofractionated regimens. Also,no experimental study has provided full four-dimensional (4D) dose reconstruction in this scenario.The authors demonstrate how a recently described motion perturbation method, with full 4D dosereconstruction, is applied to describe the gradient and interplay effects during VMAT lung SBRTtreatments.Methods: VMAT dose delivered to a moving target in a patient can be reconstructed by applyingperturbations to the treatment planning system-calculated static 3D dose. Ten SBRT patients treatedwith 6 MV VMAT beams in five fractions were selected. The target motion (motion kernel) wasapproximated by 3D rigid body translation, with the tumor centroids defined on the ten phases of the4DCT. The motion was assumed to be periodic, with the period T being an average from the empirical4DCT respiratory trace. The real observed tumor motion (total displacement ≤8 mm) was evaluatedfirst. Then, the motion range was artificially increased to 2 or 3 cm. Finally, T was increased to60 s. While not realistic, making T comparable to the delivery time elucidates if the interplay effectcan be observed. For a single fraction, the authors quantified the interplay effect as the maximumdifference in the target dosimetric indices, most importantly the near-minimum dose (D99%), betweenall possible starting phases. For the three- and five-fractions, statistical simulations were performedwhen substantial interplay was found.Results: For the motion amplitudes and periods obtained from the 4DCT, the interplay effect isnegligible (<0.2%). It is also small (0.9% average, 2.2% maximum) when the target excursion in-creased to 2–3 cm. Only with large motion and increased period (60 s) was a significant interplayeffect observed, with D99% ranging from 16% low to 17% high. The interplay effect was statisticallysignificantly lower for the three- and five-fraction statistical simulations. Overall, the gradient effectdominates the clinical situation.Conclusions: A novel method was used to reconstruct the volumetric dose to a moving tumorduring lung SBRT VMAT deliveries. With the studied planning and treatment technique for real-istic motion periods, regardless of the amplitude, the interplay has nearly no impact on the near-minimum dose. The interplay effect was observed, for study purposes only, with the period com-parable to the VMAT delivery time. © 2013 American Association of Physicists in Medicine.[http://dx.doi.org/10.1118/1.4818255]

Key words: tumor motion, interplay, VMAT, lung SBRT

1. INTRODUCTION

The challenge of tumors and other organs moving during radi-ation therapy is well documented and the history and currentstate of motion management techniques, as pertains to photontherapy, were recently reviewed by Korreman.1

The effects of motion on the tumor dose can be broadlydivided into the gradient (motion blur) and interplay effects.2

In conventional 3D conformal radiation therapy delivered inmany fractions of 1.8–2.0 Gy each, the goal is to deliver a ho-mogeneous tumorcidal dose to the target. The gradient effect,historically first noted for this type of treatment, occurs withthe occasional tumor excursion outside of the volume irradi-ated to a high, homogeneous dose. If the proper beam aper-tures are selected to encompass the entire volume potentiallyoccupied by the target, the tumor will always be irradiated to

091710-1 Med. Phys. 40 (9), September 2013 © 2013 Am. Assoc. Phys. Med. 091710-10094-2405/2013/40(9)/091710/8/$30.00

091710-2 Stambaugh et al.: Interplay in lung SBRT 091710-2

the prescribed, relatively homogeneous dose, regardless of itsmotion path within the irradiated volume. However, in abla-tive type treatments such as stereotactic body radiation ther-apy (SBRT),3 the target dose is deliberately highly inhomo-geneous, with the cooperative lung treatment protocols spec-ifying the maximum dose in the target volume between 15%and 40% higher than the prescription, RTOG 0813 (Ref. 4)being but one example. Under these circumstances, the mov-ing tumor dose will depend on the exact nature of the motion,even when the tumor does not venture outside the properlydefined target volume. This is still a purely spatial effect andit would take place regardless of whether the dose distribu-tion is produced by the static beam apertures or by a dynamicdelivery technique such as intensity modulated radiation ther-apy (IMRT) or volumetric modulated arc therapy (VMAT).However the second effect, the interplay, has a temporal com-ponent and is associated only with dynamic deliveries. Res-piratory motion can interfere with the dynamically changingbeam parameters, most notably the MLC segments’ shapes, toalter the target dose. The analogy between this process and theclassic wave behavior, including the relative frequency depen-dence and the possibility for either constructive or destructiveinterference was aptly noted.1, 5 For the homogeneously irra-diated target volume, the interplay effect is relatively easy todefine as any deviation of the tumor dose from the plannedvalue in the presence of the dynamic delivery and motion.However, with the inhomogeneous target dose distribution,distinguishing between the gradient and interplay effects in anexperiment for dynamic deliveries is more challenging. Sincethe gradient effect is purely spatial, while the interplay has atemporal component, the logical way to isolate the interplayeffect is to analyze the dosimetric effect of the varying startingphase of the motion relative to the beam start time.6

The motion effects in the presence of the IMRT deliveryhave been studied by many authors.2, 6–14 Bortfeld et al.7 pre-dicted that the interplay effect would, for the most part, aver-age out with the large number of fractions. However, it is notautomatically the case with hypofractionated radiosurgery-type course of treatment (1–5 fractions). While target motioneffects in IMRT have been studied extensively with both ex-perimental and theoretical methods, VMAT is a newer modal-ity, and so there are substantially fewer studies of interplayin VMAT,6, 9, 12, 13 particularly for hypofractionated dose reg-imens. The theoretical studies, while elucidating, do not nec-essarily rely on the actual delivery sequences, as the linearaccelerator controller may change the VMAT delivery timingcompared to the treatment planning system (TPS) estimate.15

Previous experimental studies have typically used labor-intensive physical motion stages, while reporting motion-induced dose changes to a relatively small number of points,or a single plane. In the absence of full volumetric analysis, itis not possible to evaluate the dose-volume parameters mostimportant for SBRT, such as the near-minimum target dose.

It was recently demonstrated that VMAT dose to a mov-ing target in a patient can be accurately and easily recon-structed from a single measurement in a static helical diodearray phantom, using a perturbation method,16 which incor-porates many of the advantages of both the experimental and

theoretical motion studies reported previously. Measurement-guided, motion-adjusted, high resolution (2–3 mm voxel) vol-umetric dose data on a patient are available for display andanalysis with conventional treatment planning tools. For anygiven treatment plan, only a single measurement with a staticphantom is needed to quantify the effects of any possible mo-tion pattern.

In this paper, we show how this novel methodology can beapplied to isolate and quantify the possible interplay influenceon the target dose-volume histogram (DVH) between the res-piratory motion and dynamic VMAT delivery for lung SBRTtreatments.

2. MATERIALS AND METHODS

2.A. Treatment planning and delivery

Ten consecutive patients previously treated with VMATlung SBRT were selected, starting with the most recent one atthe time of writing and working backwards. Each plan dealtwith a single lesion. Half of the patients had their tumors inthe upper lung lobes and the other half in the lower. All pa-tients were treated with 6 MV photons on Varian Trilogy orTrueBeam linear accelerators (Varian Medical Systems, PaloAlto, CA) equipped with 120-leaf Millennium MLCs (5 mmleaf width in the central region). The beam characteristicsof the different machines were closely matched (<1% dif-ferences in dose distributions). Four-dimensional helical CT(4DCT) datasets obtained on a Brilliance Big Bore scanner(Philips Medical Systems, Cleveland, OH) and reconstructedwith a 3 mm slice width were used for treatment planning.The untagged low-pitch scan was used for dose calculations.The institutional practice for VMAT lung SBRT is to use 2–3 approximately half-arcs, avoiding beam entrance from thecontralateral side. All patients were treated in five fractionsto 50 or 60 Gy prescribed to 95% of the planning target vol-ume (PTV). The internal target volume (ITV) was defined asthe union of the visible gross tumor volumes (GTVs) with noadded margins for microscopic extension, from all the dif-ferent phases of 4DCT. The PTV is derived by a uniform5 mm expansion of the ITV. Treatment planning was doneon Pinnacle treatment planning system (TPS) v. 9.2 (PhilipsRadiation Oncology Systems, Fitchburg, WI). Dose calcula-tions were performed on a 3 mm grid and 4◦ VMAT controlpoints (CP) spacing15 with the Collapsed Cone Convolutionalgorithm. For the target motion effect analysis, first the orig-inal treatment plans were used. Additionally, those plans weremodified to accommodate the expanded simulated motion, asdescribed below.

2.B. Motion kernels

2.B.1. Original empirical motion kernels

The current institutional practice is to allow VMAT treat-ments for lung SBRT tumors that move less than 1 cm. Theaverage tumor excursion was 4 ± 2 mm (1 SD), with the rangeof 1–8 mm. For the purposes of this study, the tumor motionis assumed to be a 3D rigid body translation.17 Therefore, the

Medical Physics, Vol. 40, No. 9, September 2013

091710-3 Stambaugh et al.: Interplay in lung SBRT 091710-3

FIG. 1. The moving target is the GTV drawn on Phase 0% and rigidly trans-lated according to the centroid position on each phase (two positions withmaximum separation shown). The static ITV is the union of all target vol-umes, and the PTV is the uniformly expanded ITV. The minimum and maxi-mum ITV dose voxels are shown to illustrate the gradient motion effect.

tumor motion pattern (motion kernel) can be defined as aplot of the tumor centroid position against time, discretizedin practice according to the ten 4DCT phases. The tumor wasdefined as a physician-drawn GTV on a single 4DCT phase(GTV0 by convention, Fig. 1).

The GTV0 volumes ranged from 0.9 to 17.2 cm3 [aver-age 6.8 ± 5.1 (1 SD) cm3]. Using deformable registration inthe Mirada RTx software package v.1.2 (Mirada Medical Ltd,Oxford, UK), this contour was propagated to the remain-ing nine phases. The centroids of the resulting volumes weretaken as the tumor centroid locations on every phase, provid-ing the spatial component of the original empirical motionkernel. The temporal component was based on the periodictumor motion assumption, with the average period of thebreathing trace from the 4DCT [average among the patients3.5 ± 0.8 (1 SD) s, range 2.9–4.6 s].

A union of the original tumor volumes situated on allphases following the centroid excursion represented the ITV.Since the ITVs on which the original PTVs and plans werebased were obtained by differing procedures, the tumor vol-ume was slightly adjusted so that the boundaries of the newITV were within 1 mm of the original plan ITV.

2.B.2. Simulated extended motion

Additional motion kernels simulated larger motion (2 or3 cm total target excursion). For each patient, the originalmotion trajectory was proportionately expanded to result inthe total tumor displacement of 2, and then 3 cm. Whennecessary, the direction of the motion was adjusted to avoidunrealistic patterns such as tumor incursion into the chestwall. The predominantly superior-inferior (SI) motion direc-tion was maintained. A new, expanded, ITV and PTV wereconstructed as described above, and the patients were re-planned to ensure the new PTV prescription dose coverageat the 95% volume level. Since the 2 or 3 cm tumor motionis not reflected in the planning CT densities, the ITV den-sity was manually overridden with the value of 0.8 relative

to water. The necessity and validity of this approximation areelaborated in Sec. 4.

The modified treatment plans used the same objectivesas the original ones, but additionally, the RTOG intermedi-ate dose compactness and symmetry criteria (50% dose vol-ume to PTV volume ratio, and maximum dose 2 cm from thePTV in any direction) (Ref. 4) were enforced. This resultedin higher modulation and, subsequently, dose inhomogeneityin the expanded ITVs compared to the original ones, which,if anything, should accentuate the findings of this paper. Themaximum PTV dose was 15%–40% higher than the pre-scription, conforming to the RTOG protocols specifications.4

These plans were evaluated for motion effects first with theoriginal 4DCT motion periods, and then with the 60 s period.While not realistic, making the motion period comparable tothe delivery time helps to evaluate the causes of the interplayeffect.

2.C. Data collection and analysis

2.C.1. Dose to the moving target

Measurement-guided dose estimate to a moving target ina patient is based on a research version of the commercial3DVH software package (Sun Nuclear Corp., Melbourne,FL). For the static dose distribution, the basic premise of the3DVH algorithm is that the volumetric dose on the patient canbe estimated by perturbing the patient TPS dose distributionbased on the phantom measurements.18 The dosimetry phan-tom used for VMAT beams is the helical ArcCHECK diodearray (Sun Nuclear), with its calibration formalism describedby Kozelka et al.19 The accuracy of the 3DVH static 3D dosereconstruction on an arbitrary CT dataset, from a measure-ment on a homogeneous cylindrical phantom, was demon-strated for both homogeneous18 and inhomogeneous (an-thropomorphic thoracic) (Ref. 20) phantoms representing apatient. In this study, the QA method described above wasused to confirm the agreement between planned and deliv-ered 3D dose distributions on the patients. Gamma analysiswith rather stringent 2%/2 mm (local dose-error normaliza-tion) criteria was used.

As demonstrated by Feygelman et al.,16 the planned doseperturbation concept can be extended to the dynamic (4D)dose reconstruction. The fraction dose to any patient voxel inthe presence of motion, assuming the motion kernel is known,can be derived by applying a measurement-guided motioncorrection. The dose to the diodes in a helical phantom isrecorded at 50 ms intervals and is transformed into a seriesof time-resolved high-density volumetric dose grids. A mov-ing voxel is propagated through this 4D dose space and thefraction dose to that voxel moving in the phantom is accu-mulated. For every voxel, the ratio of this motion-perturbed,reconstructed dose to the TPS dose in the phantom is taken.This ratio serves as a perturbation factor, applied to the TPSfraction dose to the similarly situated voxel in the patient.

For any given plan, the effects of multiple motion kernelscan be simulated after a single measurement. In the 3DVHsoftware, the moving region of interest (ROI) is identified and

Medical Physics, Vol. 40, No. 9, September 2013

091710-4 Stambaugh et al.: Interplay in lung SBRT 091710-4

a corresponding motion kernel (rigid 3D translation/rotation)is defined. An automatic margin equal to the diagonal of theTPS calculation grid voxel is added to the moving ROI toavoid ambiguity in scoring the border voxels as either staticor moving. It is physically justified, as the voxels close to theborder of a moving ROI are expected to move with a similartrajectory.21 A motion perturbation is applied to every voxelinside the selected ROI within the margin. The moving ROI’scentroid spatial coordinates can be typed in, or defined au-tomatically as the centroids of appropriate ROIs. The motionsimulation can be started at an arbitrarily defined time with re-spect to the beam starting moment, thus any starting phase canbe simulated. A single fraction starting at any desired phase, aspecified number of fractions with random starting phases, ora large number of fractions can be simulated in any given run.The motion-perturbed dose grid can be saved as a DICOMRT Dose object and displayed and analyzed with conventionaltools.

2.C.2. Metrics

The main metric of significance for lung SBRT is the nearminimum dose (Dmin) to the target.17 In the prior relevant pub-lications, the minimum dose encompassing 99% of the ROI(D99%) was used to represent Dmin.13, 17 On the other hand,ICRU report 83 (Ref. 22) recommends D98% as the near min-imum dose surrogate. Therefore, we investigated both andevaluated correlation between them. Similarly, D1% and D2%

were used as the near maximum dose (Dmax) metrics. Finally,the mean doses Dmean were also recorded. When necessary,the dosimetric indices for the moving target are reported asdifferences from the corresponding metrics for the static ITV,thus eliminating the influence of the fraction prescription doseand the differences in the individual plans. Throughout thepaper, we will use the term “target” to describe the movingtumor with its shape represented by GTV0. The term “ITV”will always refer to a static contour that is the union of thetarget volumes on all 4DCT phases.

2.C.3. Single and multiple fraction simulations

2.C.3.a. Single fraction. For a single fraction, the inter-play effect can be defined deterministically as the maximumdifference in the moving target dose metrics calculated forone fraction for every possible pair of the available startingphases. Our methodology allows to define and measure in-terplay according to its fundamental characteristic—the dif-ference in dose caused by the change in the motion startingphase in relation to the beam start time,2 without resorting tosurrogates.8, 10, 12 For each motion kernel with a correspondingplan, a single fraction motion simulation was run six times,corresponding to every other starting phase (0%, 20%, . . . ,80%) plus 50%, out of ten available from 4DCT. The maxi-mum spread in dosimetric indices is reported as an aggregateaverage between all patients, with corresponding ranges.

The subsequent analysis for the single fraction simulationis contingent upon the interplay magnitude. For those mo-tion kernels where the interplay effect is small, the data for

all starting phases can be averaged (i.e., interplay neglected)and, for a single fraction, the gradient effect can be quantifiedin isolation. The target dose indices are reported as differencesfrom the corresponding static ITV values. The averages andranges over the patient population are presented. When theinterplay effect is appreciable, only a combined (gradient andinterplay) motion effect can be easily discerned experimen-tally. The ranges of the target dosimetry indices in comparisonwith the ITV are presented for individual patients.

2.C.3.b. Multiple-fraction courses. Of the two motioneffects, only the interplay depends on the number offractions.7 Therefore, only when significant interplay is foundfor a single fraction is an attempt to quantify the effect offractionation warranted. Unlike for a single fraction, only astatistical estimate can be made. A simulation of three- andfive-fraction treatment courses was run five times each, withrandom starting phases assigned to each fraction for each sim-ulation. The range of dosimetry indices’ values between thedifferent runs is a measure of the interplay effect which is ex-pected to diminish as the number of fractions increases.7 Thestatistical multiple-fraction ranges for D99% were comparedto the deterministic single-fraction interplay-induced rangecaused by the different starting phases.

3. RESULTS

3.A. Single fraction

As a basic initial quality control measure, it was demon-strated that the 3D comparison between the measurement-reconstructed and TPS-calculated dose distributions with2%/2 mm gamma analysis for original plans resulted in >90%passing rate for each patient, with an average of 94.4 ± 2.7(1 SD)%. It was also determined that the changes in D99% andD98%, as well as D1% and D2% correlated very closely for allplans and motion kernels (correlation coefficient >0.99 whencompared across the patient sample). Therefore, only D99%

and D1% data are presented in detail, for clarity and brevity.In Fig. 2, the interplay effect for a single fraction is essen-

tially quantified.The maximum percentage differences in dosimetric in-

dices, between any two of the starting phases, are plottedas averages for the patient population with the correspond-ing ranges. For the original motion kernels obtained from4DCT (≤8 mm motion, 3–5 s period), the effect is negligible(<0.2%). With the motion range increased to 2 or 3 cm, themaximum difference between the phases increases slightly:averaged between the patients, the D99% interplay-inducedspread is <1%, with the maximum range of 2.2%. Therefore,for the studied plans, the interplay effect is minimal within theclinical range of motion parameters. Once the motion periodis artificially increased to 60 s, the interplay effect begins toincrease, reaching its maximum for the 3 cm motion. At thatpoint, the maximum difference in Dmin between two startingphases can be >20%.

Since for the realistic motion periods the dosimetric dif-ferences between the starting phases are small (Fig. 2), forany given patient, the moving target doses can be averaged

Medical Physics, Vol. 40, No. 9, September 2013

091710-5 Stambaugh et al.: Interplay in lung SBRT 091710-5

FIG. 2. Quantification of interplay. Maximum percentage spread of the dosi-metric indices for the moving target between any two possible starting phases.The mean values across ten patients are presented with the correspondingranges. The groups (x axis) refer, respectively, to the original motion kernelfrom 4DCT, motion range increased to 2 or 3 cm with original period, orig-inal range with 60 s period, and increased ranges with 60 s period. Withineach group, the order is D99%, Dmean, D1%.

between all the phases (i.e., interplay neglected) before com-paring to the static ITV. The results are plotted in Fig. 3, rep-resenting predominantly the gradient effect.

The differences are small for the original motion kernels.Once the motion range is increased to 2–3 cm, while main-taining the original period, the mean near-minimum target toITV dose ratios increase, while near-maximum ones decrease,characteristic of the “motion blur.” The balance of this dosechange in the current study is such that the target Dmean isalmost always higher than for the static ITV. There are onlythree instances out of possible 20 when the target Dmean foreither 2 or 3 cm motion was lower than the static ITV one.This negative difference never exceeded 1%.

For the original motion amplitude with an artificial 60 speriod, the motion-induced dose changes are still small. How-

FIG. 3. Low interplay scenario. Percentage difference between the movingtarget dosimetric indices, averaged for one fraction between all possible mo-tion starting phases, and the corresponding static ITV metrics. The data arereported as an aggregate average over all patients, with the ranges. The groups(x axis) refer, respectively, to the original motion kernel from 4DCT, motionrange increased to 2 or 3 cm with original period, and original range with60 s period. Within each group, the data order is D99%, Dmean, D1%.

FIG. 4. Hypothetical substantial interplay scenario. The motion period is60 s and the range is 2 (a) or 3 cm (b). Percentage differences are shownbetween the moving target and static ITV dosimetric indices, presented forindividual patients as a range corresponding to all possible starting phases.The groups (x axis) refer to the individual patients. Within each group, thedata order is D99%, Dmean, D1%.

ever, in comparing Fig. 3, the ranges for the original motion(<1 cm) to the original with increased period, the target near-minimum doses can now be slightly lower than for the staticITV, indicating some presence of the interplay. The maxi-mum such difference observed for a single staring phase was−3.4% for D99%.

With the 60 s period and 2–3 cm motion range, the inter-play effect results in a substantial dependence of the fractionaldose on the starting phase. Therefore, those data groups areomitted in Fig. 3. Instead, the phase-dependent spreads of thesingle fraction target to ITV dosimetric indices differences arepresented for individual patients in Fig. 4. Those results varygreatly. For example, D99% for the moving target can be any-where from 16% low to 17% high compared to the static ITV.

As an example, locations of dose-differences ≥5% [athreshold used previously by Court et al. (Ref. 9)] betweenstarting phases of 0% and 50%, with 60 s motion period and3 cm range, are presented for a single patient in Fig. 5. Nodifferences above 2% are seen for this patient with the samemotion range but with the normal period of 4.2 s.

3.B. Multiple-fraction courses

The reduction in the interplay effect with increased numberof fractions is illustrated in Fig. 6, with D99% chosen as anexample.

Medical Physics, Vol. 40, No. 9, September 2013

091710-6 Stambaugh et al.: Interplay in lung SBRT 091710-6

FIG. 5. Areas of differences above 5% for the dose distributions with 60 smotion period and 3 cm range, corresponding to the starting phases of 0% and50%. A sagittal cross-section is presented, with the contours correspondingto the maximum separation GTV positions, the ITV, and the PTV. The grayscale represents the ArcCHECK planned dose perturbation motion-blurreddose distribution obtained with 0% starting phase.

The deterministic ranges of D99% differences between themoving target and static ITV, due to the starting phase varia-tion in a single fraction, are compared to a statistical spreaddue to five simulations, with random starting phases, of thethree- and five-fraction courses. For the 2 cm, 60 s simulatedmotion, the patient-population averages of ranges in Fig. 6

FIG. 6. Reduction in interplay effect in D99% with the number of fractionspresented per patient (x-axis groups). The first data point in each group is thefull range (deterministic) between all possible starting phases of the ratio ofmoving GTV0 D99% to static ITV. The second and third ones are statisticalmeans and ranges of the same ratios between five simulations with randomstarting phases of three- and five-fraction treatments, respectively. The lastbar is the mean value for the large number of fractions (30) simulation. Themotion period is 60 s and the range is 2 (a) or 3 cm (b).

were reduced from 7.1 ± 4.2 (1 SD)% for a single frac-tion, to 3.7 ± 1.7% and 2.2 ± 1.0%, for the three- and five-fraction courses, respectively. For the 3 cm, 60 s motion, thecorresponding numbers were 8.4 ± 6.0%, 3.7 ± 1.9%, and2.8 ± 1.6%. Repeat measures analysis of variance (ANOVA)indicated that overall these differences in mean rangeswere statistically significant between the three fractionationschemes (p < 0.02). The post-test for individual pairs resultsin statistically significant differences between one and threeor five fractions at the 95% confidence level, but no statisticaldifference between three and five fractions.

The mean value for a large number of simulations (30) canalso be compared to the three- and five-fraction averages. Theresults for 2 and 3 cm motion with 60 s periods are similar.The average D99% target vs ITV differences across ten pa-tients deviated from the mean of 30 fractions by no more than0.8 ± 1.1% and 0.2 ± 1.1% for the three- and five-fractioncourses, respectively.

4. DISCUSSION

4.A. Dose to the moving target

It is clear from Fig. 2 that for the original 4DCT motionkernels (tumor excursion ≤8 mm, period 3–5 s), the interplaydefined as the maximum spread of volumetric dosimetric in-dices between all possible starting phases is negligible. Thestudied interplay increases slightly with the motion range in-crease to 2–3 cm, reaching for D99% an average value of 0.9%and the maximum of 2.2% among all the patients. A furtherslight increase is observed when the motion period is artifi-cially set to 60 s. However, even in this hypothetical scenario,the D99% is mostly higher for the target than the ITV, and isnever lower by more than 1%. With the ITV dose being al-ready higher than the prescription, the clinical effect wouldbe negligible. This coincides with the conclusions from thecomputational study by Rao et al.,13 which used largely thesame definition of interplay based on the near-minimum doseto the target. On the surface, an experimental RapidArc mo-tion study by Court et al.8 concludes that interplay can leadto appreciable changes in target dose. However, there are sub-stantial methodology differences between the current studyand that of Rao et al.13 on the one hand, and Court et al.8 onthe other hand. In this work and Ref. 13, the primary metricof interest is the near-minimum dose to the target in the pa-tient. In Ref. 8, the dose differences are recorded in a singleplane of a dosimetry array, making a DVH evaluation impos-sible. The metric was the percentage of the pixels in the tar-get, which never had dose-errors larger than a certain criterion(3%–5%), regardless of the starting phase. It can be arguedthat for the lung SBRT target dosimetry the near-minimumdose is more clinically important than the exact dose distribu-tion in the target.17

With the differences in beam modulation caused by theplanning objectives and ITV/PTV size changes when switch-ing from the original to 2–3 cm motion, it is clear fromFig. 2, that in order to observe a substantial interplay effectin the lung VMAT SBRT, two conditions must be met: the

Medical Physics, Vol. 40, No. 9, September 2013

091710-7 Stambaugh et al.: Interplay in lung SBRT 091710-7

tumor excursion must exceed 1 cm, and the motion periodmust be comparable to the gantry rotational period. With theinverse of that period being one of the fundamental frequen-cies of the delivery system, this follows the observation byKissick and Mackie5 that the maximum interference (inter-play effect) would occur when the machine fundamental mo-tion frequency is comparable to the tumor fundamental mo-tion frequency. The simulations by Li et al.11 further confirmthat the dose-error is a somewhat periodic function of both thedelivery time and motion frequency.

Approximately two-thirds of the motion kernels had pre-dominant tumor displacement in the SI direction, with the restin the AP or LR directions. At the same time, nine out of tenpatients had collimator angles between 8◦ and 15◦, and one30◦. Thus, a range of relative orientations of the leaf move-ment direction and tumor displacement14 was explored, withessentially the same results. Still, it must be emphasized thatour findings of low interplay effect should not be automati-cally extended to sites and techniques other than lung SBRT.Different optimization schemes, planning constraints, mod-ulation levels, target sizes, and motion characteristics couldlead to different results.

In Fig. 4, the near-minimum target dose tends to be higher,and the near-maximum lower than the ITV. This is due tothe bias induced by the gradient effect, which follows fromsimple geometrical considerations (Fig. 1) and represents themotion-induced averaging, or “blur.” The interplay effect issuperimposed on this gradient effect, resulting in the substan-tial range of the target dose values, depending on the start-ing phase. Interestingly, the mean target doses in Fig. 4 tendto be higher than the mean ITV dose. This manifestation ofthe gradient effect is not general but is rather a function of theintersection of the patient-specific dose distribution with thetumor position probability density function.

The quantitative comparison between the single- andmultiple-fraction interplay effects is not as rigorous becausewe are trying to compare a deterministic quantity with statis-tical samples. Only a small number of statistical simulationswere performed because of the limited practical significanceof simulating unrealistically high motion periods. However,qualitatively, the interplay effect decreased as the number offractions is increased to 3 or 5, in agreement with theoreticalsimulations.7, 11 For the data in Fig. 6, the maximum standarddeviation of D99% among the patients for five runs was re-duced from 2.9% for three fractions to 1.9% for five fractions.Averaged between all runs for all patients, the three- and five-fraction D99% did not deviate from the 30 fraction value bymore than 0.8 ± 1.1% (1 SD) and 0.2 ± 1.2%, respectively.Similarly averaged Dmean did not deviate from the 30-fractionvalue by more than 0.2 ± 0.8%, even with the unrealisticallyhigh interplay for a single fraction.

4.B. Assumptions

A number of assumptions were required to make this studyfeasible. First, the tumor motion is approximated by a rigidbody translation, as necessitated by the current 3DVH soft-ware design. This subject was studied previously by Admiraal

et al.,17 who concluded that the approximation was reasonablefor a small tumor in the lung. Second, the tumor motion is as-sumed to be periodic. If anything, the periodic nature of mo-tion is expected to exaggerate the interplay effect, and morerandom temporal motion patterns would not alter the findingsfor the realistic motion kernels (3–5 s per cycle). The 2 and 3cm simulated motion ranges can be characterized as “top ofthe range” and “exceedingly large,” respectively, for a lungtumor.23–25

Finally, we had to assign a uniform density to the ITV forextended motion (2–3 cm) simulations. Even when 4DCT in-formation is used for target definition, dose calculations areroutinely performed on an average 3D CT dataset,17, 26 whichcan be either a mathematical average of the reconstructedphases or a physical slow (low-pitch), untagged scan. Whileit results in acceptable, if not negligible, dosimetric errors forreal patient CT datasets, except perhaps at the PTV margins,27

in our case of simulated extended motion, the large portionsof the ITV not visited by a real tumor, exhibit unrealisticallylow (lung) CT density. This interferes with the perturbationalgorithm’s ability to accurately reconstruct dose to the mov-ing target. Since in this study we are interested only in relativedose changes due to the target motion within the ITV, the ex-act average ITV density is not important, as long as it is rea-sonably high. We verified that the presented numerical resultsare negligibly (<0.2%) affected by the ITV density variationsrelative to water from 0.5 to 1.0, which should encompassthe range of average ITV CT densities encountered in reallife.

5. CONCLUSIONS

We applied a motion perturbation method to reconstructthe volumetric dose to a moving tumor during lung SBRTVMAT deliveries. The possible effect of interplay betweenthe tumor motion and dynamic dose delivery was evaluatedand separated from the gradient effect that depends only onthe tumor position probability density function and 3D dosedistribution, and is present regardless of the delivery dynam-ics. With the studied planning and treatment technique, forrealistic motion periods (3–5 s), and regardless of the motionamplitude, the interplay has nearly no impact on the most im-portant target dose metric in lung SBRT—the near-minimumdose. Thus, while motion management techniques may serveuseful clinical purpose, such as reducing the amount of irra-diated healthy lung, the target dosimetry alone does not war-rant attempts to reduce the motion amplitude. The conclusionsabout the importance of the interplay can differ for differ-ent treatment sites, techniques, and optimization algorithms,while the manifestation of the effect may depend on the dataacquisition methods, analysis metrics, and the specific defini-tion of the interplay.

The interplay effect was observed, for study purposes only,with the substantial tumor motion and unrealistically long pe-riod (60 s), comparable to the VMAT delivery time. The effectof motion on the target DVH for an individual patient can befurther determined based on the experimental VMAT deliv-ery pattern recorded during a routine QA session and 4DCT

Medical Physics, Vol. 40, No. 9, September 2013

091710-8 Stambaugh et al.: Interplay in lung SBRT 091710-8

motion kernel. Future application of this method to nonrigidmotion is a matter of programming an interface that wouldtrack individual voxels through the 4D dose space in the pres-ence of organ deformation.

ACKNOWLEDGMENTS

This work was supported in part by a grant from Sun Nu-clear Corporation (SNC). B.N. is a paid consultant to SNC.

a)Author to whom correspondence should be addressed. Electronic mail:vladimir.feygelman@moffitt.org

1S. S. Korreman, “Motion in radiotherapy: Photon therapy,” Phys. Med.Biol. 57, R161–R191 (2012).

2S. B. Jiang, C. Pope, K. M. Al Jarrah, J. H. Kung, T. Bortfeld, andG. T. Chen, “An experimental investigation on intra-fractional organ mo-tion effects in lung IMRT treatments,” Phys. Med. Biol. 48, 1773–1784(2003).

3S. H. Benedict, K. M. Yenice, D. Followill, J. M. Galvin, W. Hinson,B. Kavanagh, P. Keall, M. Lovelock, S. Meeks, L. Papiez, T. Purdie,R. Sadagopan, M. C. Schell, B. Salter, D. J. Schlesinger, A. S. Shiu, T.Solberg, D. Y. Song, V. Stieber, R. Timmerman, W. A. Tome, D. Verellen,L. Wang, and F. F. Yin, “Stereotactic body radiation therapy: The report ofAAPM Task Group 101,” Med. Phys. 37, 4078–4101 (2010).

4“RTOG 0813: Seamless phase I/II study of stereotactic lung radiotherapy(SBRT) for early stage, centrally located, non-small cell lung cancer(NSCLC) in medically inoperable patients,” see http://www.rtog.org/ClinicalTrials/ProtocolTable/StudyDetails.aspx?study=0813. Accessed03/13/2013.

5M. W. Kissick and T. R. Mackie, “Task Group 76 Report on “The manage-ment of respiratory motion in radiation oncology,” [Med. Phys. 33, 3874–3900 (2006)],” Med. Phys. 36, 5721–5722 (2009).

6L. E. Court, M. Wagar, D. Ionascu, R. Berbeco, and L. Chin, “Managementof the interplay effect when using dynamic MLC sequences to treat movingtargets,” Med. Phys. 35, 1926–1931 (2008).

7T. Bortfeld, K. Jokivarsi, M. Goitein, J. Kung, and S. B. Jiang, “Effectsof intra-fraction motion on IMRT dose delivery: Statistical analysis andsimulation,” Phys. Med. Biol. 47, 2203–2220 (2002).

8L. Court, M. Wagar, R. Berbeco, A. Reisner, B. Winey, D. Schofield,D. Ionascu, A. M. Allen, R. Popple, and T. Lingos, “Evaluation of the inter-play effect when using RapidArc to treat targets moving in the craniocaudalor right-left direction,” Med. Phys. 37, 4–11 (2010).

9L. E. Court, J. Seco, X. Q. Lu, K. Ebe, C. Mayo, D. Ionascu, B. Winey,N. Giakoumakis, M. Aristophanous, R. Berbeco, J. Rottman, M. Bogdanov,D. Schofield, and T. Lingos, “Use of a realistic breathing lung phantom toevaluate dose delivery errors,” Med. Phys. 37, 5850–5857 (2010).

10E. D. Ehler, B. E. Nelms, and W. A. Tomé, “On the dose to a moving targetwhile employing different IMRT delivery mechanisms,” Radiother. Oncol.83, 49–56 (2007).

11H. Li, Y. Li, X. Zhang, X. Li, W. Liu, M. T. Gillin, and X. R. Zhu, “Dy-namically accumulated dose and 4D accumulated dose for moving tumors,”Med. Phys. 39, 7359–7367 (2012).

12C. Ong, W. F. Verbakel, J. P. Cuijpers, B. J. Slotman, and S. Senan, “Dosi-metric impact of interplay effect on RapidArc lung stereotactic treatmentdelivery,” Int. J. Radiat. Oncol., Biol., Phys. 79, 305–311 (2011).

13M. Rao, J. Wu, D. Cao, T. Wong, V. Mehta, D. Shepard, and J. Ye, “Dosi-metric impact of breathing motion in lung stereotactic body radiother-apy treatment using image-modulated radiotherapy and volumetric mod-ulated arc therapy,” Int. J. Radiat. Oncol., Biol., Phys. 83, e251–e256(2012).

14C. X. Yu, D. A. Jaffray, and J. W. Wong, “The effects of intra-fraction organmotion on the delivery of dynamic intensity modulation,” Phys. Med. Biol.43, 91–104 (1998).

15V. Feygelman, G. G. Zhang, and C. W. Stevens, “Initial dosimetric evalua-tion of SmartArc: A novel VMAT treatment planning module implementedin a multi-vendor delivery chain,” J. Appl. Clin. Med. Phys. 11, 99–116(2010).

16V. Feygelman, C. Stambaugh, G. Zhang, D. Hunt, D. Opp, T. K. Wolf,and B. E. Nelms, “Motion as a perturbation: Measurement-guided doseestimates to moving patient voxels during modulated arc deliveries,” Med.Phys. 40, 021708 (11pp.) (2013).

17M. A. Admiraal, D. Schuring, and C. W. Hurkmans, “Dose calculationsaccounting for breathing motion in stereotactic lung radiotherapy basedon 4D-CT and the internal target volume,” Radiother. Oncol. 86, 55–60(2008).

18B. E. Nelms, D. Opp, J. Robinson, T. K. Wolf, G. Zhang, E. Moros, andV. Feygelman, “VMAT QA: Measurement-guided 4D dose reconstructionon a patient,” Med. Phys. 39, 4228–4238 (2012).

19J. Kozelka, J. Robinson, B. Nelms, G. Zhang, D. Savitskij, and V.Feygelman, “Optimizing the accuracy of a helical diode array dosimeter:A comprehensive calibration methodology coupled with a novel virtual in-clinometer,” Med. Phys. 38, 5021–5032 (2011).

20D. Opp, B. E. Nelms, G. Zhang, C. Stevens, and V. Feygelman, “Validationof measurement-guided 3D VMAT dose reconstruction on a heterogeneousanthropomorphic phantom,” J. Appl. Clin. Med. Phys. 14, 70–84 (2013).

21R. L. Smith, D. Yang, A. Lee, M. L. Mayse, D. A. Low, and P. J. Parikh,“The correlation of tissue motion within the lung: Implications on fiducialbased treatments,” Med. Phys. 38, 5992–5997 (2011).

22V. Grégoire, T. R. Mackie, W. D. Neve, M. Gospodarowicz, J. A. Purdy,M. v. Herk, and A. Niemierko, “Prescribing, recording, and reportingphoton-beam intensity-modulated radiation therapy (IMRT): ICRU ReportNo. 83,” J. ICRU 10, 106 pp. (2010).

23C. W. Stevens, R. F. Munden, K. M. Forster, J. F. Kelly, Z. Liao,G. Starkschall, S. Tucker, and R. Komaki, “Respiratory-driven lung tumormotion is independent of tumor size, tumor location, and pulmonary func-tion,” Int. J. Radiat. Oncol., Biol., Phys. 51, 62–68 (2001).

24Y. Seppenwoolde, H. Shirato, K. Kitamura, S. Shimizu, M. van Herk,J. V. Lebesque, and K. Miyasaka, “Precise and real-time measurement of3D tumor motion in lung due to breathing and heartbeat, measured duringradiotherapy,” Int. J. Radiat. Oncol., Biol., Phys. 53, 822–834 (2002).

25E. D. Donnelly, P. J. Parikh, W. Lu, T. Zhao, K. Lechleiter, M. Nystrom,J. P. Hubenschmidt, D. A. Low, and J. D. Bradley, “Assessment of intrafrac-tion mediastinal and hilar lymph node movement and comparison to lungtumor motion using four-dimensional CT,” Int. J. Radiat. Oncol., Biol.,Phys. 69, 580–588 (2007).

26C. K. Glide-Hurst, G. D. Hugo, J. Liang, and D. Yan, “A simplified methodof four-dimensional dose accumulation using the mean patient density rep-resentation,” Med. Phys. 35, 5269–5277 (2008).

27M. Guckenberger, J. Wulf, G. Mueller, T. Krieger, K. Baier, M. Gabor,A. Richter, J. Wilbert, and M. Flentje, “Dose–response relationship forimage-guided stereotactic body radiotherapy of pulmonary tumors: Rele-vance of 4D dose calculation,” Int. J. Radiat. Oncol., Biol., Phys. 74, 47–54(2009).

Medical Physics, Vol. 40, No. 9, September 2013