Interplay effects in highly modulated stereotactic body radiation therapy lung cases treated with volumetric modulated arc therapy

Abstract Interplay effects in highly modulated stereotactic body radiation therapy lung cases treated with volumetric modulated arc therapy. Purpose To evaluate the influence of tumor motion on dose delivery in highly modulated stereotactic body radiotherapy (SBRT) of lung cancer using volumetric modulated arc therapy (VMAT). Methods 4D‐CT imaging data of the quasar respiratory phantom were acquired, using a GE Lightspeed 16‐slice CT scanner, while the phantom reproduced patient specific respiratory traces. Flattening filter‐free (FFF) dual‐arc VMAT treatment plans were created on the acquired images in Pinnacle3 treatment planning system. Each plan was generated with varying levels of complexity characterized by the modulation complexity score. Static and dynamic measurements were delivered to GafChromic EBT3 film inside the respiratory phantom using an Elekta Versa HD linear accelerator. The treatment prescription was 10 Gy per fraction for 5 fractions. Comparisons of the planned and delivered dose distribution were performed using Radiological Imaging Technology (RIT) software. Results For the motion amplitudes and periods studied, the interplay effect is insignificant to the GTV coverage. The mean dose deviations between the planned and delivered dose distribution never went below −2.00% and a minimum dose difference of −5.05% was observed for a single fraction. However for amplitude of 2 cm, the dose error could be as large as 20.00% near the edges of the PTV at increased levels of complexity. Additionally, the modulation complexity score showed an ability to provide information related to dose delivery. A correlation value (R) of 0.65 was observed between the complexity score and the gamma passing rate for GTV coverage. Conclusions As expected, respiratory motion effects are most evident for large amplitude respirations, complex fields, and small field margins. However, under all tested conditions target coverage was maintained.


| INTRODUCTION
Lung cancer is the second most diagnosed cancer in men and women, but the leading cause of cancer death among both. 1 Lung cancer accounts for approximately 37% of all cancer mortality and the stage of the disease at diagnosis is heavily related to outcome.
When an individual is diagnosed after the cancer has metastasized, the expected 5-year survival is approximately 4%. In contrast, when an individual is diagnosed in the localized stage of the disease the Stereotactic body radiation therapy (SBRT) is one specialized form of external beam radiation therapy treatment that utilizes hypofractionated radiation doses delivered in a limited number of fractions. Stereotactic treatments are characterized by small treatment volumes and sharp dose fall off into the surrounding healthy tissue. SBRT's hypofractionated doses have shown to be very effective at treating localized lung cancer. 8 The small treatment volumes and sharp dose fall-off limit the amount of healthy lung tissue treated; which is desirable considering that mean lung dose correlates with lung complications. 9 Although multiple studies have investigated the effects of interplay on hypofractionated SBRT treatment regimes, the results are still inconclusive on both the overall impact of interplay and the main contributors to interplay effects. One study found significant changes in target coverage for highly modulated fields and large motion amplitude 10 while another reported minimal interplay even as target excursion increased to 2-3 cm. 11 A third study reported that interplay effects would be insignificant with sufficient margin. 12 Another reported negligible interplay effects, although this study was limited by using minimal modulation in their treatment plans. 13 At our clinic, SBRT treatments have been implemented since 2009. Over the course of the program, the treatment team has observed some circumstances of high plan modulation; such as when lung tumors are in close proximity to critical structures where significant dose sparing is required. Therefore, the goal of this study is to comprehensively assess lung VMAT SBRT dose delivery under increased levels of modulation and range of motion. Evaluation of target coverage was used to gauge interplay effects on the delivered dose distributions in an effort to evaluate its clinical importance.
Gamma analysis was also used to assess the agreement between planned and delivered dose. Additionally, correlation of target coverage to a plan-based modulation metric was used to provide an indication of potentially risky plans.
In this study, a dynamic respiratory phantom (Quasar Respiratory Phantom, Modus Medical Devices, Ontario, Canada) was used as a patient model. The phantom is composed of an acrylic body and an electric drive unit was designed to simulate one-dimensional internal lung motion with various cylindrical "lung" inserts that move in the longitudinal direction while simultaneously simulating one-dimensional external chest motion with a platform that moves in the anterior-posterior direction. The phantom's acrylic body, internal and external motion features, and array of cylindrical inserts makes the device suitable for mimicking a breathing patient.
The cedar lung tumor insert was used for this study. The insert is composed of cedar wood and contains an offset, 3 cm diameter, plastic sphere to be used as lung and tumor surrogates respectively.
The CT number and density of cedar ranges between 290-400 and 0.25-0.32 g/cm 3 , which compares favorably with lung (140-300 and 0.15-0.33 g/cm 3 respectively 14 ). The CT number and density of the plastic sphere was 950 and 0.98 g/cm 3 , which compares favorably with soft tissue (1000 and 1.02 g/cm 3 , respectively 14 ). This insert, inside the phantom's acrylic body, provides the desired imaging conditions of an actual lung tumor patient. The phantom is shown in

2.A.2 | Respiratory motion models
The control software for the phantom supports import of 1D respiratory traces that can be reproduced by the phantom. Using this feature, three anonymized patient-specific breathing traces acquired with the external tracking system (Varian Real-Time Positioning (RPM), Varian Medical Systems, Palo Alto, CA) were attained. Two traces were selected where the amplitude and period of the respiratory cycles were consistent. Additionally, one irregular trace was selected where the amplitude and period of the respiratory cycles varied. The provided software utilizes a waveform editor that allows the user to filter, compress, stretch, and scale the amplitude of a given waveform. The editing software was used to scale each waveform to fixed amplitudes of 1 and 2 cm, for a total of six traces.

2.A.3 | Treatment planning image acquisition
Currently, our clinic employs a motion encompassing imaging technique for managing intrafraction respiratory motion during treatment planning and delivery of lung SBRT. This technique involves creating a region of interest (ROI) that encompasses the target and its envelope of motion. Images of the phantom were acquired with a General Electric Lightspeed multi-slice CT scanner (General Electric Company, Waukesha, WI). The parameters and values for the 4D-CT gating protocol included slice thickness of 2.5 mm, 400 mAs, 30 cm field of view, 120 kVp, and variable (5-8 s) cine duration.
The imaging protocol consists of taking an initial scout scan, a free-breathing (FB) CT scan and then a cine scan. During cine imageacquisition, the Varian Real-time Position Management (RPM) system, which consists of an infrared (IR) tracking camera and a reflective marker, monitors the external chest motion of the phantom and generates a breathing trace. Once acquired, the cine volumetric data and the RPM breathing trace were sent to the Advantage 4D-CT v1.6 binning software (GE Healthcare, Buckinghamshire, England) to create respiration-correlated CT datasets at four different breathing phases (eg full-inhale, mid-exhale, full-exhale, mid-inhale). Additionally, the Advantage software was used to create a maximum intensity projection (MIP) data set.
After imaging, the FB and MIP CT data sets were exported to a treatment planning system (TPS) (Pinnacle, 3  Each FFF-VMAT plan consisted of two full 360°treatment arcs at 6 megavoltage (MV) photon beam energy. The collimator angle was set to 45°and the couch angle was 0°. VMAT plans were generated using Pinnacle 3 SmartArc inverse planning module. 15 All plans were optimized using a 4°control points spacing and a 0.46 cm/degree leaf motion constraint. The insert is composed of cedar wood and contains an offset, 3 cm diameter, plastic sphere to be used as lung and tumor surrogates respectively. The CT number and density of cedar ranges 290-400 and 0.25-0.32 g/cm 3 , which compares favorably with lung (140-300 and 0.15-0.33 g/cm 3 respectively). The CT number and density of the plastic sphere was 950 and 0.98 g/cm 3 . calculated based on three characteristics of each segment: shape, area, and weight. Segment shape is quantified using the leaf sequence variability (LSV) parameter. LSV is defined as the variability in segment shapes of each field. The segment shape is based on the difference in leaf position between adjacent MLC leaves for each leaf bank excluding those positioned under the jaws. The maximum distance between positions for a leaf bank is defined as.

2.A.4 | Complexity score
The LSV is then calculated as follows: Segment area is quantified using aperture area variability (AAV).
AAV is defined as the variation in segment area relative to the maximum aperture area. Segments that is similar in area to the maximum aperture area contribute to a larger complexity score, ie less modulation. The AAV is calculated using the leaf position information as follows: where A is the number of leaves in the leaf bank. MCS arc is defined as follows: The PTV prescription was set to 1000 cGy per fraction for 5 fractions and the dose was calculated on a dose grid of 3x3x3 mm 3 .
This dose fractionation scheme was selected as it is allowed by the RTOG 0813 protocol, the current standard at our clinic for most cases, and common in the literature. 12,13,18 Plans were created at varying degrees of complexity characterized by the modulation complexity score. This was achieved by con- Once the kV-CBCT was acquired, it was aligned to the planning CT using automatic grey-scale matching. The designated volume for registration (clipbox) was defined to encompass the stationary parts of the phantom. Treatment plans were then delivered to the phantom, and the resulting dose was measured, with and without respiratory motion.

2.B.2 | Planar dose export
The coronal dose plane corresponding to the dose distribution measured by the film inside the phantom was exported from the TPS for each treatment plan. Planar doses were calculated for a 20 × 20 cm 2 square field. The planar dose tool, was used to create ASCII planar dose files at a resolution of 1 mm. The ASCII files were exported, and retrieved via file transfer protocol (FTP).

2.B.3 | Digitization of exposed films
Radiochromic film measurements were digitized using an Epson Expression 10000XL flatbed photo scanner (Seiko Epson Corporation, Nagano, Japan). This scanner was used to save 48-bit redgreen-blue (RGB), 150 dots per inch (DPI), images in tagged image file format (TIFF).
As recommended by the manufacturer, the films were scanned in landscape orientation to reduce lateral response artifacts. Care was taken to preserve film orientation and time between exposure and processing. Additionally, a cutout was designed to make sure that each film was placed in relatively the same position on the scanner during readout. Since the Epson scanner has no warm-up process, though 10 repeated warm-up scans were performed on the scanner before actual image digitization. Each film was digitized at 0.178 mm per pixel in order to balance resolution and document size.

2.C | Uncertainty measurements
The treatment delivery and film analysis processes are subject to error. The quality of the results presented herein is directly related to this error. Measurements were performed to quantify the uncertainty in these steps.
The uncertainty in the CBCT software registration algorithm was determined by a process similar to that described in Sutton et al. 20 The phantom was initially aligned to isocenter by a kV-CBCT using grey-value registration with only translation shifts. This gave the best alignment possible without considering rotations that the couch is unable to account for. After the phantom was aligned, six repeated measurements were acquired without moving the phantom by recalculating the registration. Ideally, these repeated measurements The quality of the film and planar dose registration process was adopted from Vinci et al. 21 The film registration software displays an estimated error value for each film registration point (δx i , δy i ) by evaluating the geometric relationship between registration points in the planar dose and film dose images. These values were taken to directly quantify the quality of the registration process, Q, as calculated by: where N = the number of registration points.
The average, range and standard deviation of Q, for N = 30, was 0.60, 0.33 -0.96, and 0.18 mm respectively. As a rule of thumb, the standard deviation (σ RIT ) should be less than or equal to 1/Pmm, where Pmm is the pixel size of the reference image in mm. For the 1 mm pixel size of our reference images, the Q value from each registration should be less than 1 mm to be considered appropriate.
This served as a quality check for all films.
Additionally, one film was registered 10 times to its corresponding planar dose distribution and evaluated. The deviation in quality (Q) from this process (σ Q = 0.03 mm) was far less than the deviations observed from registering different films. This indicated that the error is in large part due to film preparation (ie the manual cutting of film to fit inside the insert and punching of film registration holes).
Lastly, one A/P plan was generated and delivered to the phantom three times in one session to measure the end-to-end variation in the phantom setup, treatment delivery, and film analysis. Each film delivery was registered to the corresponding planar dose file in RIT.
Longitudinal and lateral profiles were acquired and the displacements between midpoints were determined using the 50% isodose line positions. This procedure includes all errors from kV-CBCT alignment, from treatment delivery, and from film registration and scanning. As measured from the data, the displacements of the midpoint at the 50% dose level between films was 1.21 mm in the longitudinal direction and 0.17 mm in the lateral direction for the film deliveries.

2.D | Analysis metrics
The various measurements and patient models examined in this work are summarized as follows: three breathing traces, seven complexities at 1 cm motion amplitude, and eight complexities at 2 cm amplitude, each plan was delivered five times. Analysis was performed on all static and dynamic film deliveries using the RIT V6.3 software package. Once the film and planned dose distributions were registered, five 1-D profile measurements along the longitudinal (superior-inferior) and lateral (right-left) axis were acquired for each film. One profile through isocenter and four profiles directly adjacent to isocenter were acquired summed and averaged.
Target coverage was evaluated using the profile measurements.
The analysis metrics evaluated the position of the measured dose distributions compared to the calculated distributions. The width of the 100% prescription dose, the width of the 95% prescription dose, and relative dose and percent dose error at the edges of the GTV, ITV, and PTV between planned and delivered dose were evaluated. Insufficient target coverage will be considered a relative dose below 95% of the prescription in the GTV and a relative dose below 90% of the prescription in the PTV. Additionally, the mean, minimum, and maximum dose difference for all points inside the GTV, ITV, and PTV were calculated between the planned and delivered distributions.
Additionally, the RIT software has the ability to perform gamma analysis. Gamma analysis 22

| RESULTS
A summary of pertinent results can be found in Table 1.

3.A | Plan monitor units
The total number of MU was recorded. The number of MU in each plan generally increased with increasing plan modulation for all three patients. Figure 2 depicts

3.B | Film analysis
Calculated dose distributions from the TPS were compared with measured static and dynamic film dose distributions via profile assessment and gamma analysis.

3.C | Profile assessment
Longitudinal and lateral profiles of each film measurement were taken. Figures 3 and 4 display profile measurements among static, dynamic, and calculated dose distributions for several cases. The plots also display the isocenter and the extent of the GTV, ITV, and PTV at the 95% prescription level (950 cGy). Lateral profiles were also taken for each film delivery. Due to the design of the insert and the dose distribution in the phantom's geometry only partial lateral profiles could be obtained. Note a lateral profile is the perpendicular to the direction of phantom motion and it is seen from each measurement that there was relatively small changes due to respiratory motion between profiles for any given amplitude and complexity. Errors were most likely due to the treatment delivery, film response, film registration, or TPS model quality.
Since the longitudinal profile corresponds to the direction of phantom motion and experiences changes in the dose distribution due to phantom motion, only the longitudinal profiles were further evaluated.
The width of the 100% (1000 cGy) and 95% (950 cGy) prescription dose along the longitudinal axis for patient trace 1 are shown in Fig. 5.
In each measurement it can be seen that the width of static dose distribution was wider than the planned dose distribution. For the 2 cm deliveries, it can be seen that the dynamic dose distributions had the shortest width and that the 95% width fails to meet the required 6 cm to cover the entire PTV as prescribed. However, all the 1 cm dynamic deliveries, meet the required 5 cm width to cover the entire PTV.

3.E | Gamma analysis
Gamma analysis was also used to assess the agreement between planned and delivered dose. Figure 6 shows RIT's gamma analysis

4.B | Limitations
One limitation to this work was the Quasar respiratory motion phantom simulates 1D motion in the longitudinal direction. Although respiratory motion is usually larger in the longitudinal direction, studies have shown that lung tumors move in all three directions. 23 Thus, it would be interesting and useful to see the effect of 3D tumor motion on the dose delivery.
As the film dosimetry process is complex, the accuracy and reproducibility of the measurement procedure could be improved.
The quality of the results is directly related to this process. This includes cutting of the film pieces to fit the insert, positioning of the film in the insert, the phantom ability to accurately reproduce the respiratory traces, etc. Additionally, another type of detector besides EBT3 film can potentially be beneficial.
Another limitation is that no statistical tests were performed on the film data because the given sample size lacked sufficient statistical power. Also, this study did not account for deformation of the tumor. Previous studies have shown that in structures such as lungs, tumor, and organ deformation can occur. 24

4.C | Future work
In the future, one could expand the number of patient specific respiratory traces to be planned, delivered, and analyzed, in order to provide a larger sample size for statistical analysis of the data. The clinical practice of SBRT is expected to increase in use for other cancers; therefore one could also expand the range of treatment sites studied. The results can be very different due to the different planning constraints, levels of modulation, target sizes, and motion characteristics. For example, the liver is another site that has been widely treated with SBRT and has considerable amount of changes in inter-fraction position due to respiratory motion. 24

CONFLI CTS OF INTEREST
The authors have no conflicts of interest to disclose.

AUTHOR CONTRIBUTION
Desmond Fernandez contributed substantially to the design, acquisition, analysis, interpretation of data, drafting and revising of the final version to be published. Justin Sick contributed substantially to the analysis, interpretation of data, drafting and revising of the final version to be published. Jonas Fontenot contributed substantially to the design, acquisition, analysis, interpretation of data, drafting and revising of the final version to be published.