Biomed. Phys. Eng. Express 5 (2019) 025043 https://doi.org/10.1088/2057-1976/ab059e

PAPER

Compensation of intrafractional motion for lung stereotactic bodyradiotherapy (SBRT) on helical TomoTherapy

Alex Price1 , Jiawei Chen2, EdwardChao3, Eric Schnarr3, Eric Schreiber2, Lan Lu2, AndreaCox3,ShaChang2 and Jun Lian2

1 Department of RadiationOncology,WashingtonUniversity in St. Louis, St. Louis,MO63110,United States of America2 Department of RadiationOncology, TheUniversity of NorthCarolina, ChapelHill, NC 27599,United States of America3 Accuray Incorporated, 1310Chesapeake Terrace, Sunnyvale, CA 94089,United States of America

E-mail: jun_lian@med.unc.edu

Keywords:TomoTherapy, intrafractionalmotion, jaw andMLC tracking

AbstractHelical TomoTherapy has unique challenges in handling intrafractionalmotion compared to aconventional LINAC. In this study, we analyzed the impact of intrafractionalmotion on cumulativedosimetry using actual patientmotion data from clinically treated patients and investigated real timejaw andmultileaf collimator (MLC) compensation approaches tominimize themotion-induced dosediscrepancy in clinically acceptable TomoTherapy lung SBRT treatments. Intrafractionalmotiontraces from eight fiducial tracking CyberKnife lung tumor treatment cases were used in this study.These cases were re-planned onTomoTherapy for SBRT, with 18 Gy×3 fractions to a planningtarget volume (PTV) defined on the breath-holdCTwithout ITV expansion. Each casewas plannedwith four different jaw settings: 1 cm static, 2.5 cm static, 2.5 cmdynamic and 5 cmdynamic. In-house4Ddose accumulation softwarewas used to compute the dose distributions with tumormotion andthen compensate for thatmotion bymodifying the original jaw andMLCpositions to track thetrajectory of the tumor. The impact ofmotion and effectiveness of compensation on the PTV coveragedepends on themotion type and plan settings. On average, the PTVV100% (the percent volume of thePTV receiving the prescription dose) accumulated from three fractions changed from96.6% (motion-free) to 83.1% (motion-included), 97.5% to 93.0%, 97.7% to 92.1%, and 98.1% to 93.7% for the 1 cmstatic jaw, 2.5 cm static jaw, 2.5 cmdynamic jaw and 5 cmdynamic jaw setting, respectively.When thejaw andMLC compensation algorithmwas engaged, the PTVV100%was restored back to 92.2%,95.9%, 96.6% and 96.4%, for the four jaw settingsmentioned above respectively. TomoTherapy lungtumor SBRT treatments using a fieldwidth of 2.5 cmor larger are less sensitive tomotion thantreatments using a 1 cm fieldwidth. For 1 cmfieldwidth plans, PTV coverage can be greatlycompromised, even over three fractions. Once themotion pattern is known, the jaw andMLCcompensation algorithm can largelyminimize the loss of PTV coverage induced by themotion.

1. Introduction

Helical TomoTherapy® (Accuray Inc., Sunnyvale, CA)is a method of radiation therapy that utilizes a helicaldelivery of dose to a patient in a slice-by-slice fashionsimilar to that of a helical computed tomography (CT)scanner [1, 2]. The gantry ring will continuously rotateas the patient continuously translates through thebore. A compact 6 MV S-band linear accelerator isattached to the gantry which delivers a fan beam that iscollimated by longitudinal jaws and a binary multileaf

collimator [1, 3]. The fan beam can laterally extend40 cm at isocenter and has a maximum longitudinalfield size of 5 cm at isocenter. In the lateral direction, abinary 64-leaf collimator is used with a leaf width of0.625 cm projected to isocenter. There are 51 projec-tions (gantry positions) per gantry rotation that areused to define the beam delivery by creating leafpatterns at each projection. Traditionally, staticlongitudinal field sizes of 1.0 cm, 2.5 cm, and 5.0 cmare used for the jaw settings. Dynamic jaws(TomoEDGE™) have been introduced that allow for

OPEN ACCESS

RECEIVED

19 September 2018

REVISED

25 January 2019

ACCEPTED FOR PUBLICATION

8 February 2019

PUBLISHED

20 February 2019

Original content from thisworkmay be used underthe terms of the CreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2019 IOPPublishing Ltd

the motion of jaws in the beginning and end oftreatment to conform to the target [4]. This techniquesallows for full field size delivery specified by the planwithin the target to speed up treatment delivery, whiledynamically closing the jaws to 1 cm at the superior-inferior borders of the target for sharper dose falloff [5].

During radiation therapy delivery, treatment tar-gets (i.e. tumors) are often not stationary. Motioninduced artifacts include dose blurring and the inter-play effect between target motion and beam modula-tion by the couch, gantry, jaws and MLC [5–11]. Theinterplay between intensity modulated deliveries andtumor motion is particularly concerning becausemany segments are smaller than the target [9, 12].Papers investigating the interplay effect quote possiblylarge dosimetric differences of up to 20%–30% for asingle fraction, but report that differences largely washout with many fractions [5, 9, 13]. A TomoTherapysimulation study by Kissick et al showed that the inter-play effect was likely small in the presence of a sinusoi-dal motion, however even a small amount ofrandomness in motion trace can change the fluenceintensity profiles dramatically [6]. Therefore, in thecontext of lung stereotactic body radiotherapy (SBRT)treatment, which uses few fractions, high dose perfraction and where the subject has realistic irregularrespiratory cycles, the interplay effect is still a concern.

The clinical impact of intrafractional motion onTomoTherapy has been reported for prostate treat-ments planned with a 2.5 cm fixed jaw plan and con-ventional fractionation [14, 15]. In the lung, smalldose blurring was observed for the delivery of Tomo-Therapy 1 cm fixed jaw plan in a phantom study, how-ever the motion used was relatively small (�1 cm) andregular [7]. There are no known publications of effectsfrom realistic intrafractional motion for the case oflung stereotactic body radiation therapy (SBRT) onTomoTherapy systems using the relatively newdynamic jaws feature.

On non-TomoTherapy systems, several intrafrac-tional motion management techniques are used clini-cally or have been studied in pre-clinical settings. Onconventional linacs, techniques including gating,breath hold and forced shallow-breathing with the useof abdomen compression have been routinely used inthe clinic [11, 16–18]. There were a series of publica-tions on integrating MLC target tracking with thefeedback from an electromagnetic position trackingsystem (Calypso 4D Localization System) [8–11].Although 1–2 mm accuracy was achieved in the exper-imental system, there is no clinical implementationtreating patients yet. The CyberKnife® TreatmentDelivery System (Accuray Inc., Sunnyvale, CA) is aclinically available radiation treatment device thatadjusts the delivery to the motion of the tumor withthe use of a mobile robotic arm. It can track the skull,spine and fiducial markers within the patient andadjust the linac head to deliver the radiation to the

target. When treating lung tumors with CyberKnife, amotion correlation model between superficial land-marks and the internal target or fiducials is built—theSynchrony® Respiratory Tracking System—and usedto dynamically guide the gantry head to follow the tar-get continuously [12].

In the case of TomoTherapy, there are fewer viableactive intrafractional motion management techniquesused in the clinic. On the research side, softwaremotioncompensation for TomoTherapy includes Motion-Adaptive Delivery (MAD) and Motion-AdaptiveOptimization (MAO) that have been previously investi-gated [13, 14, 19]. MAD is a software approach thatrearranges planning projections and leaf patterns tobest match the motion of the target. This method ofmotion compensation, however, struggles with erraticnon-periodic motion [19]. Unlike MAD, MAO per-forms single projection optimization based on thedelivered dose accumulation, motion detection andfuture dose estimation but is limited by computationalspeed [14]. Zhou et al, include pre-and-intrafractionalmegavoltage compute tomography (MVCT) images tocompensate for any changes during thefirst portionof alung SBRT treatment [20]. They found shifts were nee-ded from the intrafractionalMVCTbut did not addressconsistent intrafractional tumor motion during thetreatment. To handle real time intrafractional motionin TomoTherapy, groups have looked into gating bydelivering radiation at certain breathing amplitudepositions at specified couch positions as the patientmoves through the bore until all pulses for each anglehave been delivered [21]. There are also twomethods ofbreath hold that have been investigated: the firstmethod has the gantry move during the rest periodbetween breath-holds andwill beam-onwhenpatient isin the breath hold position [16]. The second methodscales back the dose so that the whole longitudinalextent can be covered in one breath hold and repeateduntil the prescription is reached [17].

Recently Schnarr et al [18] described the imple-mentation of sequential monoscopic imaging, and theadaptation of the motion correlation model fromCyberKnife to an experimental TomoTherapy system.Observedmotion was compensated through real-timeadjustment of the jaws and the MLC leaf patterns,which continually repoint the treatment beam at themoving target. They performed phantom studies ontheir experimental TomoTherapy machine andshowed that respiratory motion (regular and 15 mmpeak to peak) of a cylindrical target can be compen-sated in a 1 cm fixed jaw plan. The phantom plan usedin their study, however, did not have realistic patientcontours nor was the plan optimization guided byclinical objectives.

In the present study, we aim to investigate the rea-listic consequences of lung tumor motion and theeffectiveness of compensation by using real patientmotion, clinical plans optimized on actual patientanatomy and four typical jaw settings used in clinical

2

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

treatments. Unlike previous research mentionedabove, the plans used in our evaluation are clinicallyacceptable treatment plans in the SBRT lung settingfollowing in-house clinical guidelines that could beused for treatment. This approach is a further andmuch more clinical relevant investigation into thedose compensation and tracking algorithm’s ability tohandle more complex and realistic situations in thepatient treatment. As TomoTherapy works towardsproviding a clinically feasible and efficient method forreal-time tumor tracking [18], the synchronization ofintensity modulated radiation therapy (IMRT) withreal-time compensation as a method of motion man-agement needs to be further investigated. We used in-house dose calculation software [22] to explore howdose distributions are affected by target motion inTomoTherapy by including the trajectory of tumormotion.More importantly, it also provides us the abil-ity to compensate for motion by adjusting static/dynamic jaw and MLC leaf patterns in a way that issimilar to something that Accuray is developing. Thusit allows us to study this anticipated motion trackingand compensation technique when applied to realisticclinical plans and target motions, and determine whe-ther such a system could provide a competitivemethod for treating tumors in the lung SBRT setting.

2.Methods andmaterials

2.1. Treatment plans andmotion tracesEight lung cancer patients with a variety of tumormotions were included in this study. All patientsreceived two to four fiducial implants and wereoriginally treated on CyberKnife coupled with fiducialtracking. The patients were scanned with two sets ofCT scanning protocols: a high-resolution (1 mm)breath holdCT, used as the planningCT, and a normalresolution (3 mm) 4D CT. All phases of 4D CTs andplanning CT were registered on the fiducials, and thenthe residual displacement of the tumor among theseimages was used for margin estimation. The grosstumor volume (GTV) was contoured on the breathhold CT then expanded to the planning target volume(PTV) according to the margin found before.

Treatments were retrospectively planned forTomoTherapy on the breath hold CT using helicaldelivery with 18 Gy×3 fractions (54 Gy in total) to anapproximate 80% isodose line to simulate the dosime-try of a CyberKnife treatment. The PTVwas optimizedto have at least 95% of the volume to be covered by100% of prescription dose according to institutional(UNC) policy and organs at risk (OARs) were sparedper AAPM TG 101 recommendations [23]. Four dosetuning structures were contoured to force high dosesclose to the center of the GTV and quick dose fall offoutside the PTV [24]. Four clinical plans weregenerated for each patient with these commonly usedjaw settings: 1 cm fixed, 2.5 cm fixed, 2.5 cm dynamicand 5 cm dynamic. The pitch was 0.14, modulationfactors were between 1.4 to 1.8, gantry rotation periodsof 35 s to 50 s, and treatment times from about 10 minto 30 min.

The motion of the fiducials was retrieved from theSynchrony tracking model and motion log of actualCyberKnife treatments. The inter-fiducial distancewas verified by the system during the image registra-tion process before the beamwas turned on for Cyber-Knife treatment. The selected patients in this studywere all treated successfully on CyberKnife with noobvious migration of fiducials. Due to this, the fidu-cials serve as a good substitute of the tumor itself whenevaluating tumor motion. In the rest of this study, thetumor motion and fiducial motion are used inter-changeably. For each patient, the tumor motion wasrecorded over 3 separate deliveries, and each recordedmotion was unique and generally an hour long. Intra-fractional motionmanagement is neededmore for thepatients with large motion. In our previous study, wedemonstrated the motion has more of an effect alongthe longitudinal direction on the helical TomoTher-apy delivery compared to the other two directions[25]. To demonstrate the effectiveness of the proposedmotion compensation method, we purposely chose 3patients with average amplitude of peak-to-peakmotion (10–20 mm in longitudinal direction) and 5patients with very large motion (>25 mm). Table 1summarizes the tumor location, size, longitudinaldimension andmaximumperk-to-peakmotion ampl-itude along the longitudinal direction. The dominant

Table 1.Tumor location, size andmotion amplitude of sampled patients.

Tumor

location

PTV volume

(mm3)PTV longitudinal

dimension (mm)Maximum longitudinal peak to peak

motion (mm)

Fraction 1 Fraction 2 Fraction 3 Average

Patient 1 leftmiddle 16.9 35 39.3 33.9 38.9 37.4

Patient 2 left lower 12.9 31 29.6 26.4 25.3 27.1

Patient 3 rightmiddle 13.2 31 30.6 46.3 26.4 34.4

Patient 4 left lower 19.7 35 21.4 29.4 37.8 29.6

Patient 5 left lower 60.1 56 35.2 37.3 52.7 41.7

Patient 6 leftmiddle 10.8 24 8.7 21.1 14.8 14.9

Patient 7 right lower 43.4 51 11.4 18.3 17.4 15.7

Patient 8 rightmiddle 15.9 32 4.1 17.9 6.8 9.6

3

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

frequency is in the range of 0.2 Hz to 0.33 Hz. A sub-section of tumor motion data, the same length as theTomoTherapy plan delivery time (�30 min), wasselected from the middle of each recording for thedose study. The rigid motion traces of the fiducials(tumor) in all three directions were incorporated intoour dose calculation software to study the motionimpact and compensation effect.

2.2.Dose calculationwithmotionThe dose calculation software used in this research wasimplemented as a fast GPU calculation algorithm thataccurately predicts the effects of motion in TomoTher-apy dose delivery [22]. It takes targetmotion as an inputtime series and simulates a rigid translational motion ofthe target by virtually adjusting the couch to move inany direction at any time point thus replicating themotion of the tumor, then calculating the dosedeposited within that time point. The dose calculationis performed at three angles per projection and there are51 projections per rotation resulting in a dose calcul-ation resolution of one calculation about every 0.1–0.3 sfor our tested clinical plans. Jaw compensation isimplemented by shifting the jaw positions such that thecenter of the treatment beam moves with the target inthe longitudinal direction while maintaining jaw widthas specified in the original plan. In this simulation, themaximum shift of the field edge from the originalcenterline is limited to 3 cm, corresponding to thecurrent mechanical limits of the TomoTherapy jaws.Similarly, MLC compensation is achieved by adjustingthe opening of certain leaves which best match the newlocation of the target in the beams eye view transversedirection.

2.3.Workflow and data analysisThe overall suggested clinical workflow is demon-strated in figure 1. A more detailed description of theimage guided motion modelling and a dosimetricvalidation of the dose calculation software can befound in the previous phantom study papers bySchnarr et al [18] andChao et al [22], respectively. Thispaper focuses on the other aspect of the project: themotion impact on clinical plans and dosimetriceffectiveness of compensation.

In this study, we first calculated the delivered dose ineachof three fractionsusing the threeuniquemotion tra-ces for each patient. The treatment anatomy wasassumed to have not changed from the planning CT andno tissue deformation was considered. In order to com-pensate for the motion, the jaw and MLC were adjustedto track the tumor motion. Motion compensated dosefor each fraction was then re-calculated and the totaldose from all three fractions was accumulated. The per-cent volume of the PTV receiving the prescription dose(PTVV100%)was tracked for target coverage assessment.Dosimetry for lung organs at risk (OAR) was not ana-lyzed because the motion of these organs may not be

related to the recorded motions of the fiducials implan-ted in the tumor. Themotion-free original planwas usedas the reference dosimetry for comparison. Wilcoxonsigned rank test (Matlab, Mathworks, Natick, MA) wasused as statistical testing tool.

3. Results

3.1.Dosimetry of clinical plans of four jaw settingsBefore showing motion effects and compensation, wepresent the quality of the original clinical planswith nomotion (motion-free plan). A sample patient, one ofthe 8 cases studied, will help illustrate plan qualityachieved with different field width settings. Isodosesfor the sample patient, planned with a 1 cm fixed jaware shown in figure 2(a), and dose-volume histogram(DVH) curves for plans with 1 cm fixed jaw, 2.5 cmfixed jaw, 2.5 cm dynamic jaw and 5 cm dynamic jaware shown infigure 2(b). In general, the PTV is coveredby a relatively conformal dose in all of the plans exceptfor the 2.5 cm fixed jaw plan which has a less sharpdose fall-off on the superior and inferior border of thePTV compared to the other three plans due to thebigger jaw size used when treating that portion of thePTV. Most normal tissue sparing is similar, so DVHcurves for these OARs are not displayed. The specialissue with this treatment plan is that the chest wallabuts the PTV,whichmade sparing it very challenging.Our institutional policy allows minor violation of thechest wall constraint in order to strive for sufficientdose on the tumor. For the dosimetry of the chestwall, a 1 cm fixed jaw plan outperformed the others,followed by the 2.5 cmdynamic jaw plan.

Table 2 shows the statistics of dosimetric end-points of critical structures averaged for all eightpatients. As in the sample patient’s plans, the 1 cmfixed jaw plans achieve superior dosimetric quality interms of chest wall sparing and less lung dose, butsimilar quality for other important structures such asthe spinal cord and heart. The dosimetry of the 2.5 cmdynamic jaw plans was not as good as in the 1 cmplans, but generally outperformed the 5 cm dynamicjaw or 2.5 cm fixed jaw settings studied. The 2.5 cmdynamic jaw plans could be delivered in half the timeas compared to the 1 cm plans. The 5 cm dynamic jawplans generally had the least favorable dosimetry, butcould be delivered in the least amount of time.

3.2. Impact of intrafractionalmotionA 100 s selection of a respiratory trace (position versustime) of one sample patient is illustrated in figure 3(a).Please note that our treatment plans have treatmenttimes longer than 100 s and we are showing the first100 s to make the waveform in the samplemore visible.With such a respiration trace, the dose differencebetween themotion free plan and themotion impactedplan in a single fraction using a 1 cm fixed jaw iscalculated to be up to 4.5 Gy (see profile in figure 3(b)),

4

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

which is 25% of prescribed fractional dose. In thisinstance, the accumulated dose from all three fractionsdid not average out. In the accumulated results for thissample plan, the PTV is still under-dosed, and achievesonly 80.3%PTVV100%.

The change of PTV V100% after accumulating dosefrom three fractions for all 4 jaw setting variations and forall 8 patients is presented in figure 3(c). The PTV V100%

drops from themotion free plan to themotion impactedplan: the 1 cm fixed jaw plans suffered the most with anaverage deficit of 13.5% to the PTV V100% and 5 cmdynamic plans being the least affected with a deficit of

4.4%. Using the Wilcoxon signed rank test on the accu-mulated dosimetry of all plans, we found the significantchange of PTV coverage (p<0.01) for all motionimpactedplans comparedwith the original plans.

3.3. Effect ofmotion compensation by trackingBoth jaw and MLC compensation were utilizedsimultaneously to recover the dosimetry impacted bythe motion during treatment. For the sample motionreported in figure 3 and the 1 cm fixed jaw plan, thesuperior and inferior jaws were adjusted to follow themotion of the tumor along the longitudinal direction

Figure 1.The clinical workflowof intrafractionalmotionmanagement for TomoTherapy lung patient.

5

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

(figure 4(a)) over the course of approximately 1300 s(only 100 s shown for clarity). Another sample jawtrace using a 2.5 cm dynamic jaw in the scenario ofmotion tracking was demonstrated in figure 4(b) (all660 s shown). It indicates that a tight 1 cmopeningwasused on both ends of the PTV but returned to normalfield width (2.5 cm) in the middle of the PTV. TheMLC leaf pattern was shifted to adjacent leaves so thatthe open aperture was aimed at the PTV at every time

point during delivery. The TomoTherapy planningsinogram (MLC leaf pattern over time) of the originalplan, motion corrected plan and the difference areshown infigure 4(c).

The ability of the jaw and MLC to compensate formotion, along with the uncompensated motion impactof a sample patient case is shown in figures 5(a)–(d), forthe 1 cm fixed jaw, 2.5 cm fixed jaw, 2.5 cm dynamic jawand 5 cm dynamic jaw plans, respectively. The dose

Figure 2. (a) Isodose distribution of 1 cmfixed jaw clinical plan of a sample patient. (b)DVHs of PTV and the chest wall of four planswith different jaw settings for the same sample patient.

6

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

Table 2.Average of dosimetric endpoints and treatment timewithoutmotion, for plans createdwith 4 jaw settings and for all 8 patients.

Spinal cord (cGy) Esophagus (cGy) Esophagus (cGy) Heart (cGy) Heart (cc) Chest wall (cGy) Chest wall (cc) Lungs (cc)Treatment

time (s)

Constraints max dose

<2190 cGy

max dose

<2520 cGy

1770 cGy

volume<5 cc

max doe<3000 cGy 2400 cGy

volume<15 cc

max dose

<5000 cGy

3000 cGy

volume<10 cc

1160 cGy volume

<1500 cc

1 cm fixed 806.0 1210.8 0.4 1997.1 4.0 5643.9 12.2 253.8 1654

2.5 cm fixed 814.8 1242.0 0.5 2107.0 5.2 5643.8 18.1 350.0 783

2.5 cmdynamic 795.9 1223.6 0.5 2040.1 5.1 5675.5 17.5 302.7 830

5 cmdynamic 811.6 1255.1 0.7 2053.1 6.2 5634.7 19.2 323.1 595

7

Biom

ed.Phys.E

ng.Express5

(2019)025043APrice

etal

Figure 3.Motion traces and dosimetric impact of themotion. (a)A100 s selected section of themotion traces used for one samplepatient case. IEC-X is patient left-right dimension, IEC-Y is superior-inferior dimension and IEC-Z is anterior-posterior dimension.(b)Dose differences (motion impacted -motion free) and profiles of a single fraction for the sample patient case plannedwith the 1 cmfixed jaw setting. (c)The target PTVV100% ofmotion impacted plans versus original plans for all 4 jaw settings and all 8 patients afterdose is accumulated from three fractions.

8

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

Figure 4.Example jaw andMLC leaf adjustments to track the target in a sample plan. (a) First 100 s of the jaw shifts of a 1 cm fixed jawplan. (b)Complete jaw shift of a 2.5 cmdynamic jaw plan. (c)MLCadjustments. Value 1 in color scale bar represents 100% leaf opentime.

9

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

Figure 5.The dose difference ofmotion impacted plan versus the original plan (top row ofCT) andmotion compensated versus theoriginal plan (lower row ofCT) of a sample patient. PTVDVHs are on the bottom rowof eachfigure. Dose is accumulated from allthree fractions. The orange contour is the PTV. Results are shown for the (a) 1 cm fixed jaw plan, (b) 2.5 cmfixed jawplan, (c) 2.5 cmdynamic jaw plan, and (d) 5 cmdynamic jaw plan.

10

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

Figure 5. (Continued.)

11

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

difference shownon theCT (motion compensated—ori-ginal) and DVHs both come from accumulated dosi-metry of all three fractions for this sample patient. The1 cm fixed jaw plan showed more motion sensitivity.However, the compensation was capable of recoveringthe lost PTVV100% coverage for this case: the PTVV100%

without motion is 97.7%, it decreased to 80.3% withmotion, and recovered to 95.4% with motion compen-sation (figure 5(a)). For this sample patient, the planswith larger jaws settings, 2.5 cm or larger, were lessimpacted by the motion and had better dosimetricrecovery.With jawandMLC tracking, PTVV100% cover-age was restored back to 97.5%, 98.4%, and 98.9% for2.5 cm fixed jaw, 2.5 cm dynamic jaw and 5 cm dynamicjaw plan, respectively (figures 5(b)–(d)). Please note the2.5 cmdynamic and 5 cmdynamic plans both employeda 1 cm field width at the superior/inferior border of thePTV, therefore the deterioration of dosimetry at thesuperior/inferior ends of these two plans behaves simi-larly to the 1 cmfixed jawplan.However,when radiationis directed at the middle portion of the PTV, the planswith dynamic jaw settings restore the field width back tothe nominal value of the plan (2.5 cm or 5 cm). Thismade the overall dosimetry of dynamic jaw plans, asshown in the DVH curves, less severely affected by themotion and better recovered by jaw/MLC tracking ascompared to 1 cmfixed jawplan.

The benefit of motion compensation is furtherproved by the comparisons of PTV coverage between theoriginal and motion compensated plans across all plans

and all 8 patients (figure 6). PTVV100% on plans with thejaw at 2.5 cm or larger can be compensated back to closeto 95%, while the compensation on 1 cm jaw planrestored the PTV V100% to>87%. The mean dosimetryof all 8 patients given in table 3 revealed that jaw andMLC tracking recovered the average PTV V100% back to92.2%, 95.9%, 96.6% and 96.4% for 1 cm fixed jaw,2.5 cm fixed jaw, 2.5 cm dynamic jaw and 5 cm dynamicjaw, respectively.With theWilcoxon signed rank test, wefound the significant improvements of PTV V100%

(p<0.01) for three motion compensated plans (1 cmfixed jaw, 2.5 cm fixed jaw and 2.5 cm dynamic jaw)while 5 cm dynamic jaw plan is improved lesssignificantly (p=0.07).

4.Discussion

In a hypofractionation setting like SBRT, the effect ofthe motion of a lung tumor can often be substantialand not easily mitigated with fewer fractions. There-fore, in most TomoTherapy planning situations aninternal target volume (ITV) will be determined forplanning, which encompasses the maximum extent oftumormotion by aligning phase CTs on the stationarylandmarks, like the spine. This helps mitigate someconcerns of target under-dose. However, the use ofsuch an ITVwill cause higher doses to nearbyOARs, asthe treatment volume will be larger compared todirectly using aGTV toPTVexpansion on breath-holdCT. With the addition of real time motion tracking

Figure 6.The target PTVV100% of themotion compensated versus the original plans for all 4 jaw settings and all 8 patients. Dose isaccumulated from all three fractions.

Table 3.Average PTVV100% across all 8 patients, for all 4 jaw settings, and after accumulating dosefrom all 3 fractions.

1 cm fixed 2.5 cm fixed 2.5 cmdynamic 5 cmdynamic

Original 96.6±0.9 97.5±2.0 97.7±1.3 98.1±1.6Motion impacted 83.1±8.1 93.0±4.2 92.1±3.7 93.7±4.2Motion compensated 92.2±4.4 95.9±2.7 96.6±2.1 96.4±2.5

12

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

and compensation to TomoTherapy, we can insteaduse the same contours as in CyberKnife treatments,which directly employ GTV to PTV expansions. Byusing real patient motion data and clinical plans, wehave demonstrated that the proposed motion trackingmethod can largely recover PTV coverage similar tothat in the originalmotion-free planswhen using thesesmallermargins.

The 1 cm field-width plans had the best planned(no-motion) dosimetry, but they also suffered themost from motion: with motion, the PTV V100%

decreased by 13.5% on average (from 96.6% to83.1%), and with motion compensation it only par-tially recovered to 92.2% for the PTV V100% coverageon average. The other 3 field-width settings were lessaffected bymotion, where the PTVV100% decreased by5.6% or less, and was largely recovered using motioncompensation: Jaw and MLC tracking recovered thePTVV100% back to 95%or larger for plans with 2.5 cmjaw or larger. Another weakness of 1 cm jaw plans istreatment time, which is about 2–3 times longer thanthat of the other plans with larger jaw settings. So,despite the marginally better dosimetry for someOARs, the 1 cm jaw plan is a less preferred option fortreating tumors exhibiting largemotion. As a balancedsolution, a field width of 2.5 cm or larger is preferredto minimize the impact of motion and to have accep-tablemotion compensation and treatment efficiency.

Previous published studies of this jaw and MLCcompensation technique for lung motion on Tomo-Therapy presented results based on simplified targetmotions and non-clinical treatment plans [18, 22].These studies acquired film measurements to validatetheir simulation results, where both measurementsand simulations showed the motion corrected dosematched well with the planned dose (<2% difference)when the jaw and MLC tracking were engaged. In thiscurrent study, realistic motion was recorded from 24patient treatment factions and covered a much widervariety of motion patterns. Some of the motion pat-terns have a larger amplitude along with more irregu-larity such as sudden anomalies in target motioncompared to the relatively regular motion traces usedin the previous published work [18, 22]. In addition,the treatment plans included in this study also usepatient anatomy and follow a clinical protocol. Thedose falloff becomes more asymmetrical and realisticin order to spare the surrounding critical structuressuch as the chest wall and spinal cord. Overall the clin-ical plans used in this study are much more compli-cated than the previously studied non-clinical plans.Taking into account these realistic scenarios, thisstudy shows that the motion effects could be mostlycompensated on real clinical cases. It also demon-strates where this technique will struggle and whichsituations it can be beneficial to end-users within hisor her clinic. In future work, we intend to furthervalidate the motion compensation of these patientplans in a treatment delivery study.

For comparison with a prior patient study, Langenet al reported about a 0.3% drop of PTV D95% (mini-mum dose received by the hottest 95% of the volume)when delivered dosimetry of three fractions was accu-mulated for the prostate case using 2.5 cm fixed jawplan and Calypso tracking data [15]. We also use a2.5 cm fixed jaw setting; however, our lung studyshows PTV coverage as defined by the V100% can beimpaired as much as 4.5%. When a target is under-dosed, V100% and D95% index show similar trend ofdeficiency, but may scale differently dependent on theslope of the DVH curve. Secondly, this may be causedby the larger motion for the lung cases in our study,especially along the longitudinal direction where themissed dose cannot be easily remedied due to the nar-rowfieldwidth andmoving couch during treatment.

One limitation of this work is that the dosimetry ofthe normal tissue in the motion impacted and motioncompensated planswas not reported because of the diffi-culty associating the tumor motion pattern with that ofsurrounding organs. With this motion managementtechnique, fluence is moved to follow the motion of thetumor, which implies that normal tissue near the tumorshould be spared better compared to no intervention atall. In a future study, it should be possible for some dis-ease sites to estimate the dose of an OAR that is static(such as a nearby bone) or that has a knownmotion. Forexample, in the prostate case the high doses region of therectum and bladder often exists in the overlap area withtheprostate,whichmoves togetherwith theprostate. In aseparate study, Price et al showed the high dose of thebladder and rectum can be better limited in a motioncorrectedplan [25].

Another limitation of this work is the assumptionthat the patient anatomy remains the same, whenaccumulating dose over the three fractions. It has beenreported that the lung tumor size normally changesvery little among 3–5 fractions of SBRT [26]. However,for individual clinical cases the daily verification CTcould be checked to verify this.

A technical concern of tumor tracking is the accuracyof estimating real time tumor position. This concern hasbeen mostly addressed by adapting a tracking techniquefrom the established CyberKnife Synchrony technology,and by experimental study of this technique as adapted toTomoTherapy [18]. The experimental TomoTherapymachine has amounted kV imager that acquires sequen-tial monoscopic imaging to capture anatomical move-ment of the target of interest. These internalmarkers suchas fiducials can be correlated with an external breathingtrace captured by an optical camera at the foot of theTomoTherapy couch for lung targets. A model is thenbuilt that predicts the location of the fiducials based onthe prior kV images and breathingmotion trace to adjustthe delivery of the beam throughout the course of treat-ment. The model will be checked continuously againstthemost recent kV image. Themodel will be rebuilt if themodel predictiondeviates from the currentmeasurementover a threshold. This method in fact has many

13

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al

similarities with the established CyberKnife synchronytrackingmethod,whichhasbeenused to treat patients formany years [27]. The study of the prototype TomoTher-apy intrafractional management was reported in detailsby Schnar’s et al [18]. This TomoTherapy version of Syn-chrony, with motion correction enabled, produced filmmeasurements showing the dose differences in a movingtarget were less than 2%. The robustness of the positionmodeling remains to be evaluated for special cases like asudden pattern change or extreme excursion over the jawshift hardware limit. The exact latency of motion com-pensation delivery achievable in a production imple-mentation remains to be determined, although it isestimated to be about 70msecs and should not have anoticeable effect in correcting for respiratorymotion.

5. Conclusion

Our study showed that using a larger treatment beam(2.5 or 5 cm) minimizes the dosimetric impact ofintrafractional motion in TomoTherapy helicallydelivered lung tumor SBRT treatments. In addition,we showed that the ability to adjust the jaws and MLCduring treatment to dynamically compensate fortargetmotion largely eliminated the dosimetric impactof measured intrafractional motion gathered fromactual patient motion traces, regardless of whichtreatment beamwas selected during planning.

Acknowledgments

The authors appreciate Dr Michael Kissick for care-fully reviewing and editing the paper.

Conflicts of Interest

Jiawei Chen, Eric Schreiber and Jun Lian weresupported in part by a research grant from AccurayIncorporated. Ed Chao, Eric Schnarr, and Andrea Coxare Accuray employees.

ORCID iDs

Alex Price https://orcid.org/0000-0003-3135-1113Jun Lian https://orcid.org/0000-0002-2041-9074

References

[1] Mackie TR et al 1999Tomotherapy Semin. Radiat. Oncol. 9108–17

[2] Mackie TR et al 1993Tomotherapy: a new concept for thedelivery of dynamic conformal radiotherapyMed. Phys. 201709–19

[3] LangenKM et al 2010QA for helical tomotherapy: report ofthe AAPM task group 148Med. Phys. 37 4817–53

[4] Katayama S et al 2015Accelerated tomotherapy deliverywithTomoEdge technique J. Appl. Clin.Med. Phys. 16 4964

[5] ChenY,ChenQ,ChenMand LuW2011Dynamictomotherapy deliveryMed. Phys. 38 3013–24

[6] KissickMW,Boswell S A, Jeraj R andMackie TR 2005Confirmation, refinement, and extension of a study inintrafractionmotion interplay with sliding jawmotionMed.Phys. 32 2346–50

[7] KissickMW,MoX,McCall KC, Schubert LK,WesterlyDC andMackie TR 2010Aphantommodeldemonstration of tomotherapy dose painting delivery,includingmanaged respiratorymotionwithoutmotionmanagement Phys.Med. Biol. 55 2983–95

[8] McQuaidD andWebb S 2006 IMRTdelivery to amovingtarget by dynamicMLC tracking: delivery for targetsmoving intwo dimensions in the beam’s eye view Phys.Med. Biol. 514819–39

[9] Sawant A et al 2008Management of three-dimensionalintrafractionmotion through real-timeDMLC trackingMed.Phys. 35 2050–61

[10] Smith RL et al 2009 Integration of real-time internalelectromagnetic positionmonitoring coupledwith dynamicmultileaf collimator tracking: an intensity-modulatedradiation therapy feasibility study Int. J. Radiat. Oncol. Biol.Phys. 74 868–75

[11] Sawant A et al 2009Toward submillimeter accuracy in themanagement of intrafractionmotion: the integration of real-time internal positionmonitoring andmultileaf collimatortarget tracking Int. J. Radiat. Oncol. Biol. Phys. 74 575–82

[12] Dieterich S et al 2011Report of AAPMTG135: qualityassurance for robotic radiosurgeryMed. Phys. 38 2914–36

[13] LuW2008Real-timemotion-adaptive delivery (MAD)usingbinaryMLC: I. Static beam (topotherapy) deliveryPhys.Med.Biol. 53 6491–511

[14] LuW et al 2009Real-timemotion-adaptive-optimization(MAO) in TomoTherapy Phys.Med. Biol. 54 4373–98

[15] LangenKM et al 2009Dosimetric effect of prostatemotionduring helical tomotherapy Int. J. Radiat. Oncol. Biol. Phys. 741134–42

[16] Keall P J et al 2006Themanagement of respiratorymotion inradiation oncology report of AAPM task group 76Med. Phys.33 3874–900

[17] KimB,KronT, Battista J andVanDyk J 2005 Investigation ofdose homogeneity for loose helical tomotherapy delivery in thecontext of breath-hold radiation therapy Phys.Med. Biol. 502387–404

[18] Schnarr E et al 2018 Feasibility of real-timemotionmanagement with helical tomotherapyMed. Phys. 451329-1337

[19] LuW2008Real-timemotion-adaptive delivery (MAD)usingbinaryMLC: II. Rotational beam (tomotherapy) deliveryPhys.Med. Biol. 53 6491–511

[20] Zhou J et al 2010 Image-guided stereotactic body radiotherapyfor lung tumors using BodyLocwith tomotherapy:clinical implementation and set-up accuracyMed.Dosim. 3512–8

[21] KimB,Chen J, KronT andBattista J 2010 Feasibility study ofmulti-pass respiratory-gated helical tomotherapy of amovingtarget via binaryMLC closure Phys.Med. Biol. 55 6673–94

[22] ChaoEH, LucasD and Schnarr E 2018 Evaluation ofTomoTherapy dose calculations with intrafractionalmotionandmotion compensationMed. Phys. 45 18–28

[23] Benedict SH et al 2010 Stereotactic body radiation therapy: thereport of AAPM task group 101Med. Phys. 37 4078–101

[24] Soisson ET et al 2011A technique for stereotactic radiosurgerytreatment planningwith helical tomotherapyMed.Dosim. 3646–56

[25] Price A, Chao E,Chang S,Matney J,WangA andLian J 2016MO‐FG‐BRA‐07: intrafractionalmotion effect can beminimized in tomotherapy stereotactic body radiotherapy(SBRT)Med. Phys. 43 3711–3711

[26] KingK,Mayekar S,MarwahaG, Tolekidis G, CifterG andZhenH2017 Interfraction tumor size changes during lungSBRT:what factorsmatter? Int. J. Radiat. Oncol. Biol. Phys. 99E469–70

[27] Dieterich S, GreenO andBooth J 2018 SBRT targets thatmovewith respiration Phys.Med. 56 19–24

14

Biomed. Phys. Eng. Express 5 (2019) 025043 APrice et al