Dose deviations induced by respiratory motion for radiotherapy of lung tumors: Impact of CT reconstruction, plan complexity, and fraction size

Abstract A thorax phantom was used to assess radiotherapy dose deviations induced by respiratory motion of the target volume. Both intensity modulated and static, non‐modulated treatment plans were planned on CT scans of the phantom. The plans were optimized using various CT reconstructions, to investigate whether they had an impact on robustness to target motion during delivery. During irradiation, the target was programmed to simulate respiration‐induced motion of a lung tumor, using both patient‐specific and sinusoidal motion patterns in three dimensions. Dose was measured in the center of the target using an ion chamber. Differences between reference measurements with a stationary target and dynamic measurements were assessed. Possible correlations between plan complexity metrics and measured dose deviations were investigated. The maximum observed motion‐induced dose differences were 7.8% and 4.5% for single 2 Gy and 15 Gy fractions, respectively. The measurements performed with the largest target motion amplitude in the superior–inferior direction yielded the largest dosimetric deviations. For 2 Gy fractionation schemes, the summed dose deviation after 33 fractions is likely to be less than 2%. Measured motion‐induced dose deviations were significantly larger for one CT reconstruction compared to all the others. Static, non‐modulated plans showed superior robustness to target motion during delivery. Moderate correlations between the modulation complexity score applied to VMAT (MCSv) and measured dose deviations were found for 15 Gy SBRT treatment plans. Correlations between other plan complexity metrics and measured dose deviations were not found.

or even up to more than 50 mm in rare cases, are reported. The mean SI amplitudes are usually found to be below 10 mm. Motion amplitudes in the anterior-posterior (AP) and left-right (LR) directions are, not surprisingly, reported to be significantly smaller. Lower lobe tumors displacements were typically found to be larger than those of tumors elsewhere in the lung.
Dosimetric deviations due to intra-fractional motion of the target volume should be divided into blurring effects and interplay effects, as described by several authors. [6][7][8] The blurring effect will occur both for static 3D conformal radiotherapy (CRT) fields and for modulated treatment fields, and may be understood as a "smearing out" of the (more or less) inhomogeneous static dose distribution seen in the treatment planning system (TPS), due to motion. The exact dose to the moving tumor will depend on the nature of the motion, even for non-modulated fields and even when appropriate margins encompassing the entire tumor excursion are applied. For highly inhomogeneous dose distributions such as stereotactic body radiotherapy (SBRT) plans, where the central gross tumor volume (GTV) dose may be 50% higher than the peripheral planning target volume (PTV) dose, one may imagine that the blurring effect is even more prominent. The interplay effect is associated only with modulated fields such as intensity modulated radiotherapy (IMRT) or VMAT plans. Investigations on the impact of motion on dose delivery to lung tumors have been done by previous authors both by simulation and phantom measurements. Several simulation studies. 7,[9][10][11][12] have looked into modulated SBRT treatments. These reported only small dose deviations to the target volume, as long as sufficient margins were applied, and as long as tumor excursions were not too large.
One study 13 described simulation of interplay effects for both high and low doses per fraction (simulation methods were also validated with measurements). Simulated motion-induced dose deviations of up to 17% for a single fraction were reported. A number of studies have reported on the dosimetric impact of intra-fraction motion based on measurementsfor conventionally fractionated as well as hypofractionated treatment plans. Most of these used 1D or 2D sinusoidal motion, while a few reported using more complex motion such as 3D motion, or actual patient respiration curves. Reported motion amplitudes were in the 6-30 mm range. Both irradiation of phantoms with IMRT plans [14][15][16][17] and VMAT plansor a combination of VMAT and other delivery techniques. 18,19,21,29 have been investigated. The range of reported motion-induced dose deviations that is, measurements during dynamic delivery compared to measurements during stationary deliveryis quite wide. Dose deviations during non-gated delivery of up to 30%, 2-18%, and 2-5% are reported for one single field, one fraction delivered with multiple fields, and a series of 8-30 fractions, respectively.
Various authors have investigated possible correlations between plan complexity metrics and dosimetric accuracy. [23][24][25][26][27] Various measuring devices, delivery techniques, and complexity metrics were used. Results were not unambiguous, as some papers reported significant correlations, and some did not.
In radiotherapy planning, there are several options when choosing a CT image set for dose calculation. Some image reconstruction techniques and resulting image datasets may be more representative to the patient anatomy at treatment, compared to others. Other image sets may be superior in terms of accurate dosimetry, when measurements are compared to TPS calculation. 27 The various CT reconstructions as a basis for dose calculation are to some degree shown to be equivalent in terms of dosimetric accuracy and anatomical representation. [28][29][30] In the current study, we seek to determine whether optimizing and calculating plans based on different CT reconstructions has an impact on plan robustness, in terms of dosimetric deviations due to respiration-induced target motion during treatment. Relative dose deviations will be obtained by comparing dynamic measurements with stationary measurements, thereby isolating interplay and blurring effects from dosimetric agreement with TPS calculated dose. Furthermore, we seek to quantify possible interplay effects using different fractionation schemes, various motion amplitudes, and various respiration patterns. Possible correlations between measured dose deviations and various plan complexity metrics will be investigated. CT scanning procedures, contouring, and treatment planning comply with our institution's standard practice.
The novelty of the current study is twofold. To our knowledge, previous works have not investigated the possible impact that optimizing plans based on different CT reconstructions may have on robustness to target motion, in terms of measured interplay effects.
Furthermore, previous works have not, to our knowledge, investigated possible correlations between VMAT complexity metrics and measured motion-induced dose deviations for highly hypofractionated as well as normofractionated lung treatments.

2.A | CIRS phantom
A CIRS Model 008A Dynamic Thorax Phantom (CIRS Inc., VA, USA) was used in this study. This is a motion phantom suitable for investigating the impact of respiratory motion on both imaging and radiation treatment delivery for lung tumors. The phantom represents an average human thorax in shape, and contains anthropomorphic lungs and spine composed of tissue equivalent material. The linear attenuation of the lung equivalent material is within 3% of actual lung tissue attenuation for X-ray energies between 50 keV and 15 MeV. 31 The thorax phantom and its main components, illustrating the experimental setup on the Varian TrueBeam linac, are shown in Fig. 1 and motion actuator (middle) were physically connected to a cylindrical rod moving freely through one of the lung equivalent lobes of the phantom. The motion actuator allowed for precise translational motion of the rod along the couch axis, that is, the superior-inferior (SI) direction, as well as rotational motion around the same axis. The motion controller (right) was connected to the motion actuator as well as a computer, allowing remote control of the motion actuator, and thus actual motion of the rod.
The cylindrical rod is composed of the same lung tissue equivalent material, thereby making it more or less inseparable to the surrounding "lung" tissue. The rod may accommodate various inserts, such as tumor-equivalent inserts of various sizes (with or without fitted ion chamber). These inserts are placed off-axis in the cylindrical rod, enabling simulation of both anterior-posterior (AP) and left-right (LR) motion (by rotational motion of the rod) and SI motion (by translational motion along the couch axis). In this study, a spherical tumor insert with a diameter of 10 mm was used. Motion was controlled through a dedicated software, allowing independent control of the different motion axes. Motions in the AP and LR directions are to some extent dependent on each other ( Fig. 1), as rotational motion of the rod will induce simultaneous AP and LR motion. However, by modifying the start angle of the rotational motion, the relative amplitudes of AP and LR motions were modified.

2.B | Motion patterns
The phantom software includes a number of built-in motion curves.
Users may also import patient-specific curves. One may choose different motion curveswith different cycle periods and amplitudes for different motion axes, allowing simulation of complex 3D tumor motion patterns. SI motion is limited to ± 25 mm, while AP and LR motion is limited to ± 5 mm. According to the manufacturer, motion accuracy is within 0.1 mm. 31 The minimum cycle period is 1 sec, while there is no upper limit. Motion curves loop a specified number of times, and may be started and stopped at any point.
In this study, both built-in and imported patient-specific motion curves were used. The built-in cos 6 t curve simulates a respiratory motion where the exhale phase is slightly longer than the inhale phase were modified in such a way that a 70-second and a 45-second segment of the recording, respectively, repeated itself. The absolute amplitudes from the recordings were disregarded, as they represent chest wall surrogate motionnot tumor motion. The actual cycle times from the recordings, however, were regarded as actual respiration periods. The resulting motion patterns that were used during CT scanning and treatment delivery are summarized in Table 1. Amplitudes for irregular curves (P1 and P2) refer to the maximum amplitudes.  scans using GE dedicated software. The midV reconstruction was defined as the 30% phase bin (where 50% represents maximum exhale and 0%/100% represents maximum inhale). All image series were reconstructed with a slice thickness of 2.5 mm, corresponding to our current clinical practice.

2.D | Contouring
All contours were generated in the Varian Eclipse treatment planning system. Appropriate window width and window level for contouring the tumor insert was found to be 1 HU/−500 HU (Hounsfield units) based on the fact that this gave the closest agreement between the actual insert size and the resulting contour. This approach is similar to the one described by Clements et al. 32 For all three MIP reconstructions, iGTVsdefined as the volume encompassing the GTV when taking motion into accountwere generated using the above-mentioned approach. Technically, iGTV contours were generated using the thresholding option with lower and upper limits of -500 HU and 2000 HU, respectively. The three iGTV contours (different sizes due to different motion amplitudes) were copied from their respective MIP reconstruction to their corresponding exhale, inhale, AIP, and midV image series generated by motion patterns A, C, and E. This resulted in four image reconstructions sharing the same iGTV, for each of the motion patterns A, C, and Ea total of 12 series.
Thus, for each iGTV, there was an exhale image series, an inhale image series, an AIP image series, and a midV image series. iGTVs were denoted according to the motion pattern from which they were generated; iGTV A , iGTV C , and iGTV E . All iGTVs were contoured as "High accuracy structures." Clinical target volumes (CTVs) were generated by expanding the iGTVs by 5 mm, whereas PTVs were generated by expanding the CTVs by another 5 mm. This corresponds to our institution's practice for stereotactic lung treatments while it is slightly smaller than our current standard margins for curative, fractionated lung treatments.
Lungs, spinal canal, heart, and body outline contours were also generated. Three dummy "OARs" were generated to tentatively produce a higher degree of modulation during the optimization process.
Interestingly, the iGTV generated from motion pattern E, iGTV E , did not fully encompass the inhale phases. This means that our 4DCT protocol was not able to image the entire tumor motion, even with an actual patient specific motion curve and a fairly realistic amplitude. We decided to keep this underestimated iGTV due to its relevance in a clinical scenario.

2.E | Treatment planning
Treatment plans were generated in the Varian Eclipse treatment planning system, using the AAA 13.6.23 algorithm. All plans (except two static plans) were VMAT planstentatively highly modulated planned for a Varian TrueBeam linac. VMAT plans were planned with two semi arcs, and optimized using the PO 13.6.23 algorithm with heterogeneity correction enabled. Fractionation schemes were 2 Gy × 33 (normofractionated lung RT) and 15 Gy × 3 (highly hypofractionated lung SBRT). Plans with the same fractionation scheme were optimized using identical optimization parameters.
For iGTV A and iGTV C , plans were generated with the exhale phase, the inhale phase, the AIP reconstruction, and the midV reconstruction as planning CTs, respectivelyfor both fractionation schemes. For iGTV E , plans were generated with the exhale phase, the AIP reconstruction, and the midV reconstruction as planning CTs, also for both fractionation schemes. In addition, for the latter iGTV, a static, non-modulated plan with the exhale phase as the planning CT was generated for both schemes. This resulted in a total of 12 treatment plans for each fractionation scheme. Plan characteristics, in good agreement with our clinical practice for both ordinary fractionated and stereotactic plans, are summarized in Table 2.
For fractionated lung radiotherapy with a curative intent, a   For each motion pattern, plan delivery (and measurement) was initiated at three different points on the motion curves, marked by arrows in Fig. 2, simulating the random nature of treatment delivery to patients breathing freely.

2.F | Measurements
All static reference measurements, as well as the first measurement with tumor motion for every plan, were performed at least twice. A total of 143 and 128 plan deliveries were measured for 6 MV and 6 MV FFF, respectively.
The mean measured dose deviation per plan was calculated by averaging 7-15 measurements performed with the various plans.
A simplified, schematic overview of the experimental design is shown in Fig. 4. Only one "track" of the experiment is highlighted in the illustrationmotion A, the inhale CT reconstruction and the 2 Gy plan.  reportedwhich is considerably higher than those found for the hypofractionated plans in the current study, perhaps influenced by larger motion amplitudes than those used in our study. In addition, 2D measurements using radiochromic film are more likely to reveal larger dose deviations around the target borders than a centrally located ion chamber.

2.G | Plan complexity metrics
T A B L E 3 Relative dose difference for plan delivery to a moving tumor, compared to plan delivery to a static tumor, for all iGTVs and all motions patterns. Numbers are "worst case" for each motion pattern, based on at least three measurements (corresponding to three starting points on the motion curves).
iGTV CT recon. for optimization and calculation The Mann-Whitney test including Bonferroni correction showed significantly larger dose deviation for 2 Gy plans optimized with the midV reconstruction(*) compared to the three other CT reconstructions (P < 0.01). AIP, Average Intensity Projection.
The largest dose deviation measured in our study was for a single 2 Gy fraction (7.8%). The mean dose deviation per CT reconstruction and amplitude, over all measurements, was also always larger for 2 Gy plans than for 15 Gy plans. A possible explanation for this is that 15 Gy plans take more time to be deliveredmeaning that possible interplay effects are more likely to cancel out due to more respiration cycles during beam-on time. However, statistically significant differences between 2 Gy and 15 Gy plan delivery were only found for plans optimized with the midV CT reconstruction. A 2010 paper 18 reported delivery of 2 Gy plans to a thorax phantom using various delivery techniques, including VMAT. They used a tumor model moving in 3D with patient-specific motion patterns, and amplitudes comparable to the current study. Maximum motion-induced dose differences between two single fractions of up to 16% were reportedwhich is in accordance with our results for 2 Gy fractions. However, the authorsas in the current study and previous worksreport that the summed deviation after delivery of 30 fractions is typically within 2%, most likely due to averaging effects after many fractions.
In this study, the largest "worst-case" motion-induced dose deviations for 2 Gy and 15 Gy plans werenot surprisinglyfound for measurements performed with the largest maximum SI motion amplitude (motion patterns D and G, maximum SI amplitudes of 20 mm). The largest mean deviationover all measurementswas also found for motion patterns with a maximum SI amplitude of 20 mm. Thereby, it seems that patient breathing amplitude could be used at treatment planning as a parameter to identify patients who might need larger PTV margins, gated treatment, and/or special immobilizationsuch as abdominal compressionrestricting respiratory motion above a certain threshold.
Given that most lung tumors exhibit SI excursions of 10 mm or below, measurements with motion patterns A, B, E, and F are most relevant to typical clinical situations. The "worst-case" motion-induced dose deviations, for both fractionation schemes, were below 4% for measurements using these motion patterns. to some degree indicated the motion patterns of the tumorwhen disregarding absolute amplitudes. For the 15 Gy fractionation scheme, we found the largest deviation between stationary measurement and dynamic measurement when delivering a plan using iGTV C while running motion pattern G (Table 3). Motion patterns C and G had identical amplitudes, but very different motion curves ( Table 1).
The iGTV generated from motion pattern E, iGTV E , was underestimated when compared to the actual target motion during CT scanning. This implies that the target, during treatment delivery, occasionally would be outside the volume expected to encompass the entire tumor motion. However, motion-induced dose deviations for plans using iGTV E were not particularly large, even when highly irregular motion patterns were applied during delivery. The "missing part" of the iGTV was in the caudal endthe inhale phasewhere most tumors spend the least amount of time during a breathing cycle. Thus, the total time spent outside the (underestimated) iGTV constitutes a very small part of the total beam on-time. and MIt (Modulation Index total, which considers variations in gantry speed and acceleration, dose rate variation, etc.) Evaluating additional plan complexity metrics in the current study might have revealed stronger correlations between plan characteristics and the dosimetric impact of motion.
It seems there is a potential for further investigations on whether plan complexity metrics, along with the irregularity of respiratory patterns, may predict robustness of treatment plans to target motion.
A limitation of the current study is the use of only a 1-cm-diameter "tumor" sphere, which is more relevant to the SBRT setting than to the target volumes typically seen in standard, fractionated radiotherapy of lung cancer. One should also bear in mind the limitations of only measuring a single small volume in the center of the moving target. The results reported by previous authors 17 indicate that larger dose deviations are seen in the borders of the target along the direction of the motion, rather than the central part. Measuring 2D or 3D dose distributions would probably reveal such effects.
In the current study, we initiated plan delivery at three different points on the respiration curves (Fig. 2), as was done by, for example, Ong et al. 20 Using more starting points, like Jiang et al. 14 who used eight, might have yielded larger motion-induced dose deviations.

| CONCLUSION
For single 2 Gy fractions, maximum dose differences of 7.8% between static and dynamic measurements were observed. These effects appear to be attributable to interplay between MLC leaves SANDE ET AL.

ACKNOWLEDGMENTS
We acknowledge Per-Ivar Lønne for developing the Eclipse script used to calculate plan complexity metrics.

CONF LICT OF I NTEREST
No conflicts of interest.