Real‐time high spatial resolution dose verification in stereotactic motion adaptive arc radiotherapy

Abstract Purpose Radiation treatments delivered with real‐time multileaf collimator (MLC) tracking currently lack fast pretreatment or real‐time quality assurance. The purpose of this study is to test a 2D silicon detector, MagicPlate‐512 (MP512), in a complex clinical environment involving real‐time reconfiguration of the MLC leaves during target tracking. Methods MP512 was placed in the center of a solid water phantom and mounted on a motion platform used to simulate three different patient motions. Electromagnetic target tracking was implemented using the Calypso system (Varian Medical Systems, Palo Alto, CA, USA) and an MLC tracking software. A two‐arc VMAT plan was delivered and 2D dose distributions were reconstructed by MP512, EBT3 film, and the Eclipse treatment planning system (TPS). Dose maps were compared using gamma analysis with 2%/2 mm and 3%/3 mm acceptance criteria. Dose profiles were generated in sup‐inf and lateral directions to show the agreement of MP512 to EBT3 and to highlight the efficacy of the MLC tracking system in mitigating the effect of the simulated patient motion. Results Using a 3%/3 mm acceptance criterion for 2D gamma analysis, MP512 to EBT3 film agreement was 99% and MP512 to TPS agreement was 100%. For a 2%/2 mm criterion, the agreement was 95% and 98%, respectively. Full width at half maximum and 80%/20% penumbral width of the MP512 and EBT3 dose profiles agreed within 1 mm and 0.5 mm, respectively. Patient motion increased the measured dose profile penumbral width by nearly 2 mm (with respect to the no‐motion case); however, the MLC tracking strategy was able to mitigate 80% of this effect. Conclusions MP512 is capable of high spatial resolution 2D dose reconstruction during adaptive MLC tracking, including arc deliveries. It shows potential as an effective tool for 2D small field dosimetry and pretreatment quality assurance for MLC tracking modalities. These results provide confidence that detector‐based pretreatment dosimetry is clinically feasible despite fast real‐time MLC reconfigurations.


| INTRODUCTION
Stereotactic body radiation therapy (SBRT), stereotactic ablative radiotherapy (SABR), and stereotactic radiosurgery (SRS) are highly conformal radiotherapy modalities for small tumors and functional abnormalities. They use a high dose per fraction (from 3 to 20 Gy) and fewer fractions than conventional external beam radiotherapy (EBRT) over the course of treatment. Field sizes are also typically smaller than conventional radiotherapy, commonly less than 4 9 4 cm 2 , where effects such as the loss of charged particle equilibrium (CPE) and partial source occlusion are present.
SABR is often used to treat early-stage lung tumors. 1 Patient respiration has been reported to affect target position by up to 30 mm during EBRT. 2 Intrafraction tumor motion can have a considerable impact on the dose delivered to the tumor volume and surrounding healthy tissue, 3 if motion is not compensated. Careful quality assurance processes are essential in the application of such highly conformal radiotherapy modalities when tumor motion is present. 4 Intrafraction tumor motion can be accommodated in the treatment planning process using an extra margin called the internal target volume (ITV). This margin encompasses the entire volume occupied by the moving tumor throughout several breathing cycles, obtained from 4DCT data. Inclusion of an ITV generally leads to a larger planning target volume (PTV) and as a result, there is often more healthy tissue irradiated compared to a treatment experiencing minimal tumor motion. Another approach to manage organ motion is gating the beam to treat only when the tumor is in a particular position. [5][6][7] Real-time adaptive radiotherapy (ART) aims to treat the target volume with sensible margin reduction, in comparison to an ITV contouring, by monitoring the position and shape of the tumor continuously throughout the treatment in order to modify the beam in real time. 8 The aim is to reduce the size of the ITV considerably, so that healthy tissue receives minimal dose, while at the same time delivering the correct dose to the target to achieve the highest possible therapeutic outcome.
Kilovoltage Intrafraction Monitoring (KIM) is one real-time monitoring technique 9 ; it uses the gantry equipped kV x-ray imager to monitor the position of the tumor in real time during treatment. An algorithm calculates the difference in tumor position from the current and previous image to predict the target position in 3D. When the tumor moves outside a predefined tolerance, the treatment is paused and the patient is repositioned through use of couch motion prior to resuming treatment.
There are several other possible strategies to manage patient motion during EBRT, including pretreatment and intratreatment imaging, respiratory control devices or robotic couches.
Recently, an extensive review was published of current motion management strategies applied in the clinic. 10 An alternative motion management approach in real-time ART uses Calypso, the electromagnetic guided tumor tracking adopted by Varian, and real-time beam modulation using the multileaf collimator (MLC). The Calypso system detects and reconstructs the position of electromagnetic beacons implanted in proximity of the tumor volume. 11,12 The combination of Calypso and MLC tracking has been implemented clinically at Royal North Shore Hospital (Sydney-Australia) as a real-time tumor monitoring solution for ART. In 2013, MLC tracking was used for the first time during a prostate VMAT treatment 13 and more recently with lung SABR. 14 In that study, the dose distribution was recalculated on the patient CT dataset using a motion encoded treatment plan derived from analysis of the treatment log files (dynalog). Results showed a reduction in the PTV size by up to 40% and a reduction of mean dose to normal lung tissue of 30%, compared to a standard ITV-based delivery. 14 Tracking techniques such as KIM and radiofrequency tracking use feedback systems that may induce a signal in the chosen dosimeter, i.e., kV x rays and RF electromagnetic radiation, respectively. It is important that the dosimeter used during real-time ART is also unperturbed by these additional signals and only measures the dose due to the MV treatment.
Small radiation fields introduce some extra dosimetry requirements compared to larger fields. Namely high spatial resolution of the detector array for use in steep dose gradients as well as small sensitive volume of a single detector, compared to beam size, to avoid volume averaging. The adopted dosimeter system should ideally be close to tissue equivalent, so as not to perturb the beam. 15,16 There are several dosimeters that can potentially be used for small field dosimetry, each with their advantages and disadvantages. EBT3 film and Fricke or polymer-based gel have the required spatial resolution for small field dosimetry. They can give 2D and 3D dose information, respectively; however, they lack a real-time readout. 16,17 Ionization chambers can be read out in real time and can be calibrated for absolute dose; however, their spatial resolution and volume averaging effects can limit their use in high-resolution arrays for small field dosimetry. 16 Silicon diodes can be fabricated with a small sensitive volume; this means they can form 2D arrays with high spatial resolution. The density of silicon is considerably higher than that of water, so energy dependence needs to be characterized as well as dose rate and angular dependence. 18 Log file analysis is a reasonable machine QA tool for MLC and gantry positional errors. However, it is unable to provide information on the position of a moving phantom relative to the beam or independently evaluate the performance of an external tracking system, such as Calypso.
The Centre for Medical Radiation Physics (CMRP) at University of Wollongong has developed a 2D monolithic silicon diode array detector to be used for QA of small field real-time ART: MagicPlate-512 (MP512). It has been designed to independently monitor external beam radiotherapy treatments under complex clinical conditions and reconstruct dose distributions in real time. It is capable of pulseby-pulse monitoring of the linac beam and has a spatial resolution suitable for small field dosimetry. It has been rigorously characterized for basic square fields in terms of percentage depth dose, output factor, beam profiles, dose per pulse, and dose linearity. 19 It has also been used with modulated IMRT fields, with a fixed gantry angle and a single patient motion applied. 20 This study shows dosimetric results from MP512 using a two-arc VMAT delivery with realistic patient motions and tracking by the means of the Calypso localization system. Dose was delivered to a solid water phantom for three different patient motion traces obtained by 4DCT at Royal North Shore Hospital (Sydney-Australia). Dose maps obtained from MP512 were compared to EBT3 film and treatment planning system calculations.  Each diode has a sensitive area of 0.5 9 0.5 mm 2 with 2 mm center-to-center distance. The detector is assembled on a 500 lm thick printed circuit board which allows for mechanical rigidity and connection of the readout electronics.

| MATERIALS AND METHODS
MP512 is a monolithic silicon detector which has an intrinsic asymmetric structure creating an angular dependence, which needs to be considered. Angular correction factors for 6 MV photons on a Varian 2100iX linac have previously been obtained. 21 The correction is valid for all MP512 devices; individual diode sensitivity is corrected pixel-by-pixel by a flat field prior to measurements. The rotation of the linac was independently monitored by an inclinometer positioned on the gantry and synchronized with the data acquisition system. This allowed for on-line correction of the detector response frame-by-frame.
To calibrate the detector, a dose linearity measurement in the dose range 50 cGy to 1000 cGy was delivered at d max in solid water for a 10 9 10 cm 2 field size with 6 MV photons from a Varian 2100iX linac. Detector charge (nC) was plotted against dose (cGy) and the slope of the linear relationship was used to define the dose calibration factor.

2.B | Reproducing organ motion
The patient lung motions were simulated using the 6D Hexamotion  These motion traces were chosen as they represent three unique types of respiration. Figure 2 shows the motion traces as a function of time for each patient in x, y, and z directions for approximately 80 s, corresponding to the time required to deliver the planned dose to the target.

2.C | Motion tracking
MLC tracking is performed using the Calypso 4D localization system (Varian Medical Systems -Palo Alto, CA, USA) for real-time monitoring of the target motion in conjunction with an MLC driving software developed at University of Sydney. 23 Electromagnetic beacons are used to determine the target position, this information is used by the tracking software to signal the MLC controller to configure the field shape in response to the detected motion of the target. 13 Two different algorithms were input to the MLC controller: passive feedback and predictive feedback. The passive feedback algorithm calculates the position of the moving target based on information from Calypso, and instructs the MLC controller to reconfigure the beam aperture accordingly. There is a measurable latency of approximately 230 ms 13 between detection by Calypso and actuation of the MLC leaves due to computation time and finite leaf speed, meaning that the beam lags behind the real target trajectory. The predictive feedback algorithm uses kernel density estimation 23

2.D | Detector encapsulation and phantom
MP512 was encased between two 5 mm thick sheets of PMMA for structural rigidity (see Figure 1). There is an air gap above the sensitive volume to match the detector response to that of EBT3 film for small field beams down to 1 9 1 cm 19,25 and to minimize the density and perturbation effects of silicon. 26,27 An aluminum shielding box of 2 mm thickness was designed to mitigate the baseline fluctuations due to the electromagnetic (EM) field generated by Calypso. 20 The time-varying EM field induced a current in the wiring and DAQ electronics, which contributed to fluctuations of the baseline signal (noise). The aluminum box covered and sealed the entire detector and data acquisition system. Its effect on dosimetry has been investigated by the means of Geant4 and proved experimentally to have negligible effects at 1.5 cm water equivalent depth and below. 20 The phantom geometry is such that the minimum thickness of solid water in the beams eye view is  Table 1.

2.E | Treatment planning and delivery
Two VMAT arcs were used to deliver 496 MU and 508 MU at Note that the same plan was delivered for all motion modalities, i.e., there was no ITV-based plan used for comparison. The aim of this work was to characterize MP512 for use in adaptive arc modalities using MLC tracking and gantry rotation. There was no intention to comment on the effectiveness of MLC tracking versus ITV-based planning in terms of clinical outcomes; hence, the same plan was used for all modalities.
In Figure 4, a slice of the CT dataset incorporating the GTV and PTV margins is shown. Figure 5 depicts the 2D dose map from Eclipse overlaid on MP512 from a beam's eye view (coronal plane).
The square regions overlaid on Figure 5 highlight the sensitive area of the detector and correspond to the dose map position extracted for comparison with the experimental results. The baseline noise of the detector signal was determined to be approximately AE1%, the dose uncertainty of the EBT3 film was calculated 19 to be approximately AE2% and error bars are reported on the dose profiles, accordingly.

3.A | Dose linearity
The response of MP512 to accumulated dose is shown in Figure 6.
The relationship of charge to dose in the 50 to 1000 cGy range is linear. The MP512 measured charge is converted to dose using slope of the linearity curve.

3.B | Gamma analysis
The integral dose maps obtained from EBT3, MP512, and TPS are shown in Figure 7. Angular corrections were applied frame-by-frame to the MP512 response before integrating the dose.
The gamma pass rates comparing EBT3 to MP512 and TPS to MP512 for the no-motion case are shown in Table 2. Table 3 shows the gamma analysis results for the three motion patterns in respect to the no-motion case as measured by MP512.

3.C | Dose profiles measured by MP512 and EBT3
film In Figure   modalities. The uncertainty in EBT3 film measurement was calculated 19 to be AE2% and is within the size of the data points. Analysis of the baseline fluctuation of MP512 produces an uncertainty of AE1%. Table 4 summarizes the analysis of the FWHM and penumbral width for the dose profiles for the sup-inf direction.

3.D | MP512 motion profiles
The profiles shown in Figure 9 highlight  be used as a tool to quantify the efficacy of a particular tracking regime, and in patient-specific pretreatment QA.

| DISCUSSION
Gamma analysis was also performed on the MP512 measured dose maps to intercompare each motion modality, this gives some insight into the effect of MLC tracking.
Patient 1 motion (with no tracking) resulted in a pass rate of 78% (2%/2 mm); however, the two tracking modalities brought the pass rate to 95% or above for both 3%/3 mm and 2%/2 mm acceptance criteria. Patient 2 uncorrected motion had a pass rate of 55% (2%/2 mm) and Patient 3 had 80% pass rate (2%/2 mm) with motion. When tracking was applied, pass rates of above 93% were observed for both Patient 2 and Patient 3 for both criteria.
Gamma pass rates above 95% for 3%/3mm agreement are commonly accepted as sufficient to proceed with treatment. 28  This information can be used to give insight into the efficacy of a particular tracking regime. For example, the results from the gamma analysis show the predictive tracking algorithm did not always lead to a higher pass rate when compared to passive tracking, in this case it was particularly dependent on the type of motion being experienced by the target volume.
In our study, for periodic motion patterns, the predictive algorithm had superior performance and gave better target tracking; however, for motion patterns that are aperiodic and erratic the passive tracking gave better results. We hypothesize this is due to the difficulty in accurately predicting an irregular signal and depends on the type and duration of the learning process adopted in the predictive algorithm. 31 Some algorithms use an initial learning window before treatment where the program learns the breathing trace over a few respiratory cycles and others employ a dynamic type of learning which can adapt faster to irregularities in the motion. 32 Dose profiles from MP512 and EBT3 film undergoing Patient 1 motion were plotted on a single axis and compared ( Figure 8) found that in the sup-inf direction (major axis of organ motion due to the breathing cycle), the penumbral width of both profiles was within 0.5 mm and FWHM agrees within 1 mm for all deliveries.
The agreement of data between MP512 and EBT3 film over multiple datasets further proves the accuracy of MP512 for use in adaptive therapy treatments.  however, there is an overdose of 8% on the central axis in this direction when no tracking is applied which is reduced slightly to around 6% when tracking is activated.
The shift of the dose in the horizontal direction is evident in this case due to the large lateral component of the motion trace.
Although the MP512 sensitive area is not large enough to reconstruct the full dose profile in this direction, the effect of motion causing distortion of the profile exemplifies the benefits of tracking.

| CONCLUSIONS
CMRP has developed MP512, a 2D silicon diode array for use in small field dosimetry. It consists of 512 individual ion implanted diodes of size 0.5 9 0.5 mm 2 with 2 mm center-to-center distance.
In this study, a two-arc VMAT treatment was delivered to the phantom and MP512's angular dependence was corrected frame-byframe. MP512 response was comparable to EBT3 film and the calculated TPS dose. These results provide confidence that detectorbased pretreatment dosimetry is clinically feasible despite fast realtime MLC reconfigurations.