Experimental and Monte Carlo characterization of a dynamic collimation system prototype for pencil beam scanning proton therapy

Abstract

Purpose:

There has been a growing interest in the development of energy-specific collimators for low-energy pencil beam scanning (PBS) to reduce the lateral penumbra. One particular device that has been the focus of several recent published works is the dynamic collimation system (DCS), which provides energy-specific collimation by intercepting the scanned proton beam as it nears to target edge with a set of orthogonal trimmer blades. While several computational studies have shown that this dynamic collimator can provide additional healthy tissue sparing, there has not been any rigorous experimental work to benchmark the theoretical models used in these initial studies. Therefore, it was the purpose of this work to demonstrate an experimental method that could integrate an experimental prototype with a clinical PBS system and benchmark the Monte Carlo methods that have been used to model the DCS.

Methods:

An experimental DCS prototype was designed and built in house to actively collimate individual proton beamlets during PBS within a well-characterized experimental setup. Monte Carlo methods were initially used to assess construction tolerances and later benchmarked against measurements, including integral depth dose and lateral asymmetric beamlet profiles. The experimental apparatus and measurement geometry were modeled using MCNP6 benchmarked from measurements performed at the Northwestern Chicago Proton Center.

Results:

Gamma analysis tests were used to evaluate the agreement between the measured and simulated profiles with a strict 1 mm/1% criteria and 5% dose threshold. Excellent agreement was observed between the simulated and measured profiles, which included 1 mm/1% gamma analysis pass rates of at least 100% and 95% for the integral depth dose (IDD) profiles and lateral profiles, respectively. Differences in the relative profile shape were observed experimentally between beamlets collimated on- and off-axis, which was attributed to the partial transmission of the beam through an unfocused collimator. Exposure rates resulting from the activation of the device were monitored with survey meter measurements and were found to agree with Monte Carlo estimates of the exposure rate to within 20%.

Conclusion:

A DCS prototype was constructed and integrated into a clinical dose delivery system. While the results of this work are not exhaustive, they demonstrate the effects of beam source divergence, device activation, and beamlet deflection during scanning, which were found to be successfully modeled using Monte Carlo methods and experimentally benchmarked. Excellent agreement was achieved between the simulated and measured lateral spot profiles of collimated beamlets delivered on- and off-axis in PBS. The Monte Carlo models adequately predicted the measured elevated plateau region in the integral depth-dose profiles from the low-energy scatter off the collimators.

1. INTRODUCTION

The dynamic collimation system (DCS) is an emerging technology to be incorporated into low-energy pencil beam scanning proton therapy to provide energy-specific collimation through the sequenced motion of four collimation trimmer blades.¹ These trimmer blades are capable of rapid, independent motion just above the patient surface, intercepting the scanned proton beam as it nears the target periphery. There exists a substantial amount of literature that focuses on the potential benefits of the DCS device among head and neck and intracranial treatments delivered with both the IBA Universal and Dedicated Nozzle systems^2–4 and also in comparison to single-field aperture collimated treatments.⁵ These computational studies have utilized an asymmetric beamlet model from Gelover et al.⁶ to perform efficient dose calculations. The difficulty in modeling asymmetric proton beamlets is reflected in the complexity of the model, which was parameterized from linearly regressing numerous polynomials over energy-dependent Gaussian parameters to approximate Monte Carlo-simulated beamlet profiles. This model has been integrated within a research-based treatment planning system and has led to several published works on the DCS, including the development of novel treatment planning optimization techniques and applications.^7,8

The application of energy-specific collimators in PBS can improve the target conformity. Consequently, the precision of a treatment delivered with the DCS will be dependent on the accuracy of the calculation models used to plan the treatment. As there lacks any consensus standard of dosimetry for these small-field proton treatments, Monte Carlo transport methods are heavily relied on to calculate dose distributions⁹ and assess plan robustness. While Monte Carlo is a useful tool, its results are dependent on the geometric and physics-related assumptions made by the user and need to be experimentally benchmarked.^10–15 These assumptions include how the incident energy distribution, beam divergence, multiple Coulomb scattering energy loss, and large angular scatter are modeled, which influences the expected beam range, lateral profiles, distributions in heterogeneous matter, and absolute dose differences from changes within the incident proton intensity.

The integration of the DCS into a clinical environment will require extensive benchmarking and inter-comparisons among measured, calculated, and simulated dose distributions. To date, research associated with this technology has remained theoretical with limited experimental benchmarking of Monte Carlo methods. This has remained a difficult feat since a clinical prototype of the system is still under development. The redundancy of these metrics is rather important to consider as the consequences of error in setup or calculation are greater due to the high target conformity expected from this technology. Therefore, it was the purpose of this work to design and construct an experimental version of the DCS prototype that could easily be integrated into a clinical dose delivery system in order to provide an experimental benchmark of Monte Carlo calculation models of the asymmetric beamlets generated using a DCS device.

2. MATERIALS AND METHODS

2.A. Experimental dynamic collimation system prototype

The DCS was designed using SolidWorks^® (Version 2017) and constructed in-house. The experimental prototype, illustrated in Fig. 1, was modeled to closely mimic the Monte Carlo geometry used by Hyer et al.1 and Gelover et al.⁶ to initially study the applications of asymmetric beamlets. For many proton therapy systems, the lowest deliverable beam energy remains too penetrating for low-energy treatment sites such as head and neck and peripheral brain tumors. As such, an external range shifter is often necessary to adequately cover the target and was integrated within the DCS design so as to maximize the benefits of collimation during low-energy treatments. While several features in the DCS design are meant to be clinically adaptable, it was not the intent for the prototype to serve in any clinical capacity, including beamline QA or patient treatment. The focus of the experimental design was to provide:

FIG. 1.

SolidWorks^® model of the experimental dynamic collimation system from the front (a). Power and ethernet cables were run through the grooves within the Rexroth 30-series framing toward the power source and router. A visual rendering of the simplified Monte Carlo model of the prototype and simulation geometry (b) used to simulate primary proton and secondary activation rates resulting from irradiation.

1. Lightweight, compact proton collimator that can be easily transported, but large enough to collimate beamlets up to 7.5 cm off-axis

2. Inexpensive component cost and construction resources

3. Operable using a single computer remotely from a control vault

4. Modular design that can be used across different proton therapy beam lines

5. Reproducible and repeatable collimator translation and alignment

6. Well-defined geometry easily reproduced in Monte Carlo simulation codes

7. Constructed within tolerances that would result in acceptable errors

8. Compatibility with scanning water tanks and two-dimensional (2D) scintillation plate dosimeters

Two 20.32 cm × 3.00 cm × 2.80 cm trimmer blades were milled from nickel-200 alloy (Corrosion Materials, Inc., Bolingbrook, IL). Nickel-200 is a commercially pure, low-alloy nickel that has a high density and low secondary neutron production rate making it a promising material for collimator design in particle therapies. Each trimmer blade was secured using a stainless steel alloy-304 mating plate to a high-precision Velmex 8.0 inch XSlide^™ chassis assembly and stepper motor. The motion stage was then attached to a 30.0 cm × 30.0 cm rectangular frame constructed from Resxroth^® aluminum 6063-alloy 30-series profiles (Bosch Corp., Lohr am Main, Germany). A front face was laser cut from 0.25 inch stock acrylic to hold a 20.0 cm × 20.0 cm, 7.5 cm²=g thick range shifter fabricated from high-density (1.82 g/cm³), AR-12 grade iso-molded graphite (Ohio Carbon Blank, Inc., Willoughby, Ohio). Iso-statically compressed calcine petroleum are known for their uniform distribution of graphite grains and pours that lack a grain direction. Grain uniformity is a particularly important trait for range-shifting applications in proton beam therapy as variations within the local grain deposits can cause a spatially non-uniform range degradation across the face of the range shifter. Therefore, several tests were later performed benchmarking the resulting water-equivalent thickness (WET) uniformity of the carbon range shifter.

Strict machining tolerances were used to minimize the influence of geometrical differences between the Monte Carlo model and physical measurements. Parts were nominally machined to design within a 0.05 mm precision. The trimmer’s rotational alignment normal to the frame base was aligned to within 0.05 mm along the width of the trimmer while the pitch and yaw of the trimmers were aligned to within 0.25 mm along the length of the trimmer with respect to the frame base as illustrated in Fig. 2.

FIG. 2.

SolidWorks^® model downstream illustration of the collimation trimmers and base frame referencing points during the construction and component alignment. Base leveling (red), trimmer rotation (green), and pitch (cyan) used the lower corner of the DCS as a common reference point. Trimmer yaw (yellow) was evaluated along the entire length of the trimmer

2.B. Dose delivery framework and beam line

Measurements of collimated beamlets from a DCS were performed at the Northwestern Proton Center (Warrenville, IL). All measurements were completed in a horizontal orientation using the fixed beam line fitted with the Ion Beam Applications (IBA) Universal Nozzle that was PBS enabled. Custom scanning patterns and spot deliveries were achieved through user-defined PBS layer delivery (PLD) files. Each file describes the following beam characteristics: the energy layers delivered, the number of beam spots, the position of all beam spots in addition to the total intensity weight of the delivered field, individual energy layers, and individual spots. The delivery of treatments defined by the PLD files were ran in the RMS standby mode at the console. Multiple input files can be written if a specific order of energy layers or beam spots are desired.

Software was written using C# from the AxisClassLibrary supplied from the manufacturer and ran using VisualStudio (Version 7.7.4, Microsoft Corp.) to translate the trimmer blades, monitor their positions along their axis of motion, and home the motors remotely. The motors were powered with a 24 V power source with 5 V logic circuits to monitor limits switches placed on the extreme end-of-line travel along the Velmex motion track.

2.C. Alignment and delivery protocols

Profiles resulting from the DCS were measured using either a three dimensional (3D)-scanning water tank or 2D-scintillation plate. The DCS prototype was placed upstream of isocenter with the treatment couch placed distally to the collimator while supporting the measurement apparatus. The base of the frame was offset 1.0 cm from the dose delivery system’s isocenter using a high-precision ruler along the treatment couch. Brass shims were placed underneath the base of the DCS frame to horizontally level the device using a magnetic analog level placed on either the X- or Y-trimmer blade. The DCS was aligned to within 0.5 mm of the desired longitudinal displacement from isocenter and to within 0.5° in any rotational degree of freedom.

The coordinate systems defining the trimmers’ axis of motion and the dose delivery coordinate system were co-registered after the initial alignment. A coarse alignment was performed by centering the trimmer blades to the alignment lasers mounted in the treatment vault. Then 400 monitor units (MU) were then delivered in a 10 cm × 10 cm uniform field of 128.5 MeV, 2.5 mm laterally spaced proton beamlets to fine tune the registration. The resulting lateral profile was measured using a Lynx scintillator (IBA, Belgium) placed 1 cm downstream of the base frame so as to minimize the effects of magnification when calculating trimmer alignment offsets. As the collimators block a portion of the irradiated protons, a set of rectangular band profiles were used to measure the full-width at half maximum across the delivered field and the portion of the field blocked by the collimators, effectively averaging over several pixelized line profiles. As illustrated in Fig. 3, the offset of these regions provides the necessary translation of the trimmers to coincide with the center of the delivered field. This iterative process was automated in MATLAB (version R2019a, MathWorks, Nitack, MA) and repeated until the measured offsets of the trimmers were within 0.25 mm of the uniform field, which is within the resolution of the Lynx detector. Following this alignment, the known offset was set using a system of limit switches that could be referenced to reset the collimator positions during measurements.

FIG. 3.

Visual rendering of the alignment profile measured using the Lynx scintillator (left). Software written in MATLAB was created to measure the width of the penumbra blocked by the trimmers, shown as gray rectangles, relative to the field (plotted right). This was used to determine a shift in the trimmer’s placement to align the origin of the trimmer’s axis of motion to the origin of the PBS system.

2.D. Characterization measurements

The water equivalent thickness (WET) of the range shifter was measured using a range pull-back technique¹⁶ with a 171.32 MeV isocentric spread out Bragg peak proton beam. Depth dose measurements were acquired using an IBA Zebra multi-layer ionization chamber (MLIC, IBA), which was used to determine the depth displacement of the distal 80% penumbra with and without the range shifter placed in front of the MLIC. In order to characterize the spatial uniformity of the range shifter, several measurements of the WET were made with the range shifter laterally displaced in front of the MLIC.

Integral depth dose (IDD) and lateral profile measurements were acquired as a basis to provide a well-benchmarked Monte Carlo model of the DCS prototype and evaluate the reliability of the Monte Carlo calculations to predict changes in profiles due to the influences of scatter from the collimators. The experimental setups are illustrated in Fig. 4. An IBA StingRay chamber, with a sensitive collecting diameter of 12 cm, was used to measure the IDD profiles acquired horizontally through the wall of an IBA Blue Phantom² 3D Scanning water tank. A 151.0 MeV proton beam was instantiated with a steady current of 20 nA at the origin. Ion chamber measurements were acquired in 1 mm depth increments for isolated and coherent X- and Y-trimmer positions of 0.15, 0.50, 1.00, and 2.50 cm from the central axis. Similarly, lateral profile measurements were acquired using the IBA Lynx scintillator. Agreement between the measured with the Monte Carlo-simulated profiles were evaluated using a 1%/1 mm gamma criteria.

FIG. 4.

Experimental setup for lateral profile measurements with the IBA Lynx scintillator (left) and integral depth dose measurements with the IBA Blue Phantom² and StingRay chamber (right) of individually collimated proton pencil beams generated with the experimental dynamic collimation system prototype.

The change in beamlet intensity was quantified from the summed signal within the 50% beamlet penumbra. Beam spot centroid displacement resulting from collimation, relative to an untrimmed beam spot delivered at the same spatial coordinate, was evaluated from the center of the 95% isodose beamlet centroid. These metrics are illustrated in Fig. 5.

FIG. 5.

Demonstration of spot centroid shift due to collimation. The X- and Y-trimmers were placed 0.15 cm from the origin for the trimmed spot. The total shift was determined from the distance between the untrimmed and trimmed centroids approximated by the center of their respective 95% isodose region (highlighted in red).

2.E. Benchmarking the Lynx scintillator in non-standard conditions

The IBA Lynx^® scintillator is composed of a 30 cm × 30 cm gadolinium-based scintillation screen and a high-resolution CCD camera. The scintillation light is reflected from a mirror and collected by the photodiodes within the CCD camera. After collecting the signal, various corrections are internally applied to account for image deformation, thermal and communication electronic noise. Up to a 0.5 mm spatial resolution is achievable near the center of the detector.

Existing studies have demonstrated the signal stability and linearity across the manufacturer-recommended range of iris settings, between 20% and 100%.^14,17 However, the generalizability of these results is limited as these studies only investigated one nominal beam energy at the beam isocenter measured in air. There is little published work that has evaluated the response of this detector outside of these reference conditions and was necessary for this work given the presence of the external range shifter mounted on the DCS. Therefore, a secondary check was performed to first benchmark the lateral profile measurements of the Lynx to the profiles measured with Gafchromic^® EBT3 film. While EBT3 film has been shown to have a non-linear sensitivity quenching response for high LET radiations,^18,19 previous work has shown the quenching effect to be constant with change in optical density and thus a suitable dosimeter for relative dosimetry.²⁰

The dose response of the Lynx scintillator was benchmarked across multiple beam energies and residual depths in Gammex Solid Water^®. For each irradiation, a 10 cm × 10 cm uniform field was delivered with beam spots placed 2.5 mm apart at the isocenter. All beam spots within the field were delivered with the same number of MUs. These broad-field irradiations were planned using the RayStation treatment planning system (Raysearch Labs, Sweden) to determine the central axis depth dose profile per MU. Pencil beam scanning delivery files were generated from these treatment plans and delivered at the console to achieve a specified dose at the measurement depth. The average scintillation signal was determined within a small ROI within the center of the uniform profile. The signal linearity of the Lynx scintillator was assessed from a trust-region optimized linear fit performed using MATLAB’s fit function and reporting the maximum residual between measured and modeled dose response.

2.F. Monte Carlo modeling

Monte Carlo N-Particle Transport code version 6 (MCNP6)²¹ was chosen for all simulations in this work as it, and its predecessor MCNPX, have been shown as an effective tool to model PBS proton systems^22,23 in addition to maintaining continuity of this experimental work with earlier theoretical studies.^1,6,24 Beam energies were assumed to be Gaussian-distributed about a mean nominal energy and were matched so that the distal Bragg peaks coincided with commissioned IDD profiles. The lateral profiles were mathematically back-projected to an angularly divergent beam source distribution from the measured lateral profiles of each beam energy benchmarked during commissioning.

Simulations were performed to establish machining tolerances that were used during the construction of the experimental dynamic collimation system and test the sensitivity of alignment procedures developed to co-register the collimation trimmers translational axis with the beam scanning axis. However, the focus of this work was to benchmark radiation transport methods to accurately model asymmetric proton beamlets from a DCS. Unlike previous studies that used a simplified DCS, a representative model of the primary components of the DCS were included, such as the trimmers, trimmer holders, translation assemblies, frame, front face, and range shifter as shown in Fig. 1. The simulation geometry was modeled to reflect the experimental setups for both the IDD and lateral profile measurements. The transport parameters were also fine-tuned from these earlier studies to include a finer sampling of energy loss tables, efac = 0.955, and ion recoil sampling, recoil = 3. Nuclear interactions were included using the mix-and-match option to first use nuclear cross sections when defined and then to allow physical models in MCNP6 when measured data are not available. Range straggling was modeled with Vavilov scattering theory.

An initial set of simulations were performed to determine the influence of component misalignment on the resulting IDD and lateral profiles using 128 and 173 MeV proton beamlets incident on the trimmers and range shifter. Changes in the relative longitudinal displacement up to 2 mm in magnitude and 2° of deflection were simulated using a surface transformation card in MCNP6 as illustrated in Fig. 6. Type 3 energy deposition mesh tallies were used to score dose for the depth dose and lateral profile simulations. Integral depth dose profiles were scored using cylindrical tallies with a 6.0 cm radius, which was equivalent to the diameter of the StingRay Bragg peak chamber used during measurements, and spaced 1 mm apart in-depth. Approximately 1E6 histories for each beam energy were simulated to achieve relative simulation uncertainties of less than 0.5% at the Bragg peak. Lateral profiles were scored in 0.5 mm voxels at the center of a 4 mm-thick slab of water to mimic the WET to the plane of detection measured by the Lynx scintillation plate. A total of 1E7 histories were run for each lateral profile resulting in simulation uncertainties of less than 3% at the peak. Each measured and simulated lateral profile was post-processed using an equivalent 2D Gaussian smooth function in MATLAB.

FIG. 6.

The three degrees of rotation used to study the impact of collimator misalignment with respect to the incident beamline.

Device activation was also modeled to estimate the exposure per delivered MU and as a function of beam current. The problem was simplified by simulating a single 173 MeV proton beamlet directed toward the center of the DCS. During the simulation the trimmers were centered along each axis to calculate the resulting proton and neutron fluence from a completely absorbed proton beamlet within the device. The tallied spectra was then used to estimate the activation rates of various unstable nuclei from a compiled library of nearly 100 interaction cross sections acquired from the TENDL libraries.²⁵ A separate set of simulations were then performed to determine the exposure rate by tallying the air collision kerma in a spherical shell with a radius of 1 m from the center of the device using a *F8 pulse height tally. Positron annihilation and resulting 511 keV photons were assumed to be uniformly emitted within a 3 cm × 3 cm column within the center of the range shifter and trimmers weighted by the ratio of isotope production in each component. A single exposure measurement was taken following the series of measurements using the internal Geiger counter from a Thermo Scientific^™ Radiameter and are compared to the respective time point predicted using the Monte Carlo methods.

3. RESULTS

3.A. Machining tolerances

IDD and lateral profiles resulting from a 170 MeV incident beamlet collimated by the DCS prototype were simulated in water with the X- and Y-trimmers 0.15 mm off-axis. The collimators and range shifter were translated longitudinally and rotated from their initial alignment along each degree of rotation: rotation, pitch, and yaw as illustrated in Fig. 6. The resulting change in IDD and lateral profiles are shown in Fig. 7 for multiple rotations of the trimmer blade. No significant effect was observed for the magnitude of trimmer pitch offsets studied in this work. Rather than a single line profile across the X-axis, yaw deviations from the incident alignment were compared using the differences between the 2D profile distributions plotted in Fig. 8.

FIG. 7.

The resulting simulated lateral profiles across the x-axis (left) and in depth (right) from a 173 MeV proton beamlet in water collimated with the trimmer blade undergoing a rotation transformation as illustrated in Fig. 6.

FIG. 8.

Lateral profile effects from a 2° (left) and 0.5° (right) yaw rotation (Δ Yaw) on the x-trimmer collimating a 128 MeV proton beamlet tallied at the surface. Differences are shown in percent of the profile maximum from a beamlet collimated with an undeflected trimmer (Nominal). Isodose lines for each profile are overlaid from the 90% down to the 10% and 1%.

3.B. Depth dose measurements

Measured IDD profiles were shifted downstream to account for the WET of the Blue Phantom² wall, the chamber wall, 3.5 cm chamber alignment offset from the wall of the water tank, and the WET of the range shifter. The simulated and measured WET of the range shifter were 67.2 mm and 67.1 mm ± 0.1 mm, respectively. Beam spot profiles are particularly sensitive to the distance a trimmer is located relative to the spot’s central axis. Therefore, the positioning accuracy of the delivered beamlet was accounted for by applying small offsets to the simulated beam spot’s lateral position relative to a baseline alignment measurement so as to properly compare measured and simulated spot profiles. The offset from the intended spot position delivered with the PBS system was determined from the most X- and Y-trimmer collimated profiles, which resulted in the simulated beamlet to match the measured IDD profile within a 1%/1 mm gamma criteria. This offset was fixed for all other simulated collimated and un-collimated IDDs plotted in Fig. 9. It was assumed that the beam spot did not move in position after a steady beam current was established. The subsequent simulated and measured IDDs matched 100% with the aforementioned gamma criteria.

FIG. 9.

StingRay-measured and MCNP6-simulated integral depth dose distributions of all trimmer configurations weighted to StingRay maximum signal (top row), single depth dose curve with comparative gamma-test results when the X trimmer was set 0.15 mm off-axis (bottom left), and the elevated plateau region depth dose predicted from Monte Carlo²¹ and benchmarked via measurement with and without trimmers.

3.C. Lateral profile measurements

EBT3 film was initially used to benchmark the lateral profiles measured from the Lynx scintillator along the central axis. Prior to the profile measurements, the Lynx scintillator’s dose response was characterized among several LET environments by placing the face of the detector in increasing WET depths using Gammex Solid Water^®. The signal linearity to increasing doses was assessed by the percent difference of the maximum residual following a linear fit optimized using a trust-region method default to MATLAB’s fit function. The relative profile from the EBT3 film was converted from net-ΔOD following the methods of Smith et al.²⁰ and is plotted in Fig. 10.

FIG. 10.

Comparison of lateral profiles measured between film and Lynx scintillator (left) and the linearity of the Lynx scintillator assessed for three water-equivalent depths with increasing proton LET (right). Error bars represent the standard deviation of pixel values for each measurement.

Lateral profiles were measured and simulated along the center and 5 cm off-axis in the x- and y-direction. Following the Gaussian smoothing, the noise among pixels in both the Monte Carlo- and Lynx-measured profiles were on the order of 0.5% in the high-dose region. A global 1%/1 mm gamma test was used to evaluate the agreement between simulated and measured profiles. Several profiles are plotted in Fig. 11 for the 151.0 MeV beamlet delivered along the central axis. Output differences were assessed from each profile by summing the raw signal defined within each beamlet’s 50% penumbra and normalizing the signal to the respective uncollimated case, which are listed in Table I. Centroid shifts were calculated by the change in the beamlet’s centroid center of mass position defined within the 95% penumbra. These results are listed among the Monte Carlo and measured profiles in Table II.

FIG. 11.

Simulated (solid line) and measured (dotted line) lateral profiles from beamlets delivered and collimated along the central axis using a 151.0 MeV proton beam. Gamma index maps with a 1%/1 mm gamma criteria are shown on the right for the respective profiles on the left. The gamma passing rate is listed in the upper left-hand corner for each profile. Each column depicts the profile from a given offset a trimmer is placed from the central axis. The trimmers used to collimate the beamlet are indicated along each row.

TABLE I.

Outputs of collimated beamlets defined by the integral signal within the 50% isodose line normalized to the output of an uncollimated proton beamlet.

		128.5 MeV			151.0 MeV
Trimmer (X or Y)	Offset (cm)	Lynx	MC	Diff (%)	Lynx	MC	Diff (%)
X	0.15	0.609	0.611	−0.2	0.628	0.635	0.7
	0.50	0.851	0.851	0	0.902	0.896	0.6
	1.00	0.992	0.991	0.1	1.001	1.001	0
Y	0.15	0.633	0.624	0.9	0.689	0.653	3.6
	0.50	0.876	0.853	2.3	0.925	0.897	2.8
	1.00	0.989	0.986	0.3	1.001	0.997	0.4
X & Y	0.15	0.383	0.380	0.3	0.430	0.415	1.5
	0.50	0.732	0.723	0.9	0.832	0.803	2.9
	1.00	0.980	0.976	0.4	1.000	0.997	0.3

Results and differences are shown for those simulated using Monte Carlo methods (MC) and measured using a Lynx scintillator and are normalized to the maximum signal from an uncollimated beamlet.

TABLE II.

Measured and simulated centroid displacements (in cm) from the un-collimated beamlet centroid position defined by the 95% isodose region.

		128.5 MeV			151.0 MeV
Trimmer (X or Y)	Offset (cm)	Lynx	MC	Diff (cm)	Lynx	MC	Diff (cm)
X	0.15	0.14	0.14	0	0.13	0.13	0
	0.50	0.09	0.10	−0.01	0.07	0.07	0
	1.00	0.03	0.03	0	0	0.02	−0.02
Y	0.15	0.15	0.15	0	0.13	0.12	0.01
	0.50	0.09	0.10	−0.01	0.08	0.07	0.01
	1.00	0.02	0.03	−0.01	0.01	0.03	−0.02
X & Y	0.15	0.20	0.21	−0.01	0.18	0.17	0.01
	0.50	0.12	0.13	−0.01	0.10	0.09	0.01
	1.00	0.03	0.04	−0.01	0.02	0.03	−0.01

Distances are given in cm between Monte Carlo (MC) simulation and measurement with a Lynx scintillator.

The impact of beamlet deflection on a beamlet’s lateral penumbra while using a non-focused collimator was evaluated among the trimmed beamlets delivered along the central axis and those delivered off-axis. Figure 12 shows the lateral profiles from a beamlet trimmed with the Y-trimmer on-axis and displaced 5 cm (about 1.25° scanning deflection) from the y-axis. Simulated and measured relative profiles were overlaid so that their peak regions coincide. In each scenario, the low-dose penumbra tail toward the collimator was elevated between 1% and 2%.

FIG. 12.

Simulated and measured lateral profiles from a 128.5 MeV X- and Y-trimmed beamlet delivered on and off axis. The collimating trimmer was displaced 0.15 cm from the nominal spot position at the isocenter whereas the other trimmer was placed at least 2.5 cm from the spot position in order to not affect the resulting dose distribution. The profiles measured from the off-axis beamlets have been shifted back by their nominal displacement from the origin.

3.D. Exposure rate estimates

The activation from discrete MU deliveries was assumed instantaneous to simplify the modeling of the induced activity following a series of lateral profile measurements. Since the IDD measurements used a sustained beam, the calculations were adjusted to account for an MU rate rather than discrete MU deliveries. A sequence of the DCS’s exposure rate following these studies is shown in Fig. 13. Device alignment scans required 400 MU while each individual measurement with the Lynx was performed with 20 MU. The agreement between the post-exposure measurement following the each series of measurements are compared to the Monte Carlo methods in Table III.

FIG. 13.

Modeled device activation and resulting exposure rate following a night’s measurement of IDD profiles with a constant beam current (top) and lateral profiles assuming discrete delivery of MU increments delivered to the device (bottom). The device exposure was measured approximately 1 m from the center and 10 cm from the surface of the experimental DCS following IDD and lateral profile measurements, respectively.

TABLE III.

Measured and simulated exposure 10 cm in front of the experimental DCS prototype following a night of IDD and profile measurements (top and bottom row, respectively).

		Measured		Simulated		MU delivered
IDD measurements		20 mR/h		16.4 mR/h		2.21E7 MU
Lateral profile measurements		40 mR/h		37.1 mR/h		2.80E3 MU
Uncertainty (%)	Type A	Type B			Type A	Type B
Signal stability	5.77		Proton Stat. error		0.13
137Cs calibration		3.25	Proton transport			5.22
Scale uniformity		5.77	Kerma Stat. error		2.34
Energy dependence		1	Photon transport			0.19
			Detector position			7.5
Combined k = 1	8.8		Combined k = 1		9.4

Measurements were taken 20 and 90 min following a night’s IDD and profile measurements, respectively. The uncertainty is approximated for both measured and simulated exposure rates.

The measured exposure rate uncertainty includes the contribution from a ¹³⁷Cs calibration offered at the UWRCL and a conservative estimate of the scale rate dependence assumed rectangularly distributed 10% about the mean. The signal stability was also observed to follow a rectangular distribution about the mean exposure value. The dominant uncertainty in the Monte Carlo exposure rate estimates was the inverse-squared relationship of the effective source-to-detector distance approximated from the measurement. While statistical errors in the simulation were small, the uncertainties in the proton nuclear cross sections and MCS tables are more significant. However, this trend was flipped for the simulation of kerma in a 1 cm diameter volume placed in-air above the range shifter. Both simulated and measured exposure rates were within each other’s k = 2 confidence interval.

4. DISCUSSION

This work provides an initial experimental benchmark of the theoretical models used to study the clinical applications of a DCS in addition to providing a continuity between the initial Monte Carlo studies and future clinical prototypes currently under development. It was not the intent of this work to design nor implement a clinically usable prototype, but to demonstrate broadly its feasibility and methods of characterization. In this regard, the simplicity of the device was advantageous; as an external collimator, it was easily to implement and align so that the measurement geometry was easily modeled and reproducible. As a result, little error in the geometric modeling is expected between measurement and Monte Carlo simulations.

4.A. Machining tolerances

Tight machining and alignment tolerances are recommended following the initial Monte Carlo simulations on component alignment. Little-to-no difference was observed for lateral and depth dose profiles resulting from longitudinal displacements simulated up to 1 mm of the range shifter and trimmers. Lateral and depth dose profiles were insensitive from pitch offsets within the collimators. In an instance of a 2° rotation, the path length through the trimmer blade and range shifter increases by 0.02 mm and 0.03 mm, respectively. Rotation of the trimmer, so that the medial edge is no longer parallel to the beamlet, had a profound effect on the lateral and depth dose distributions. Rotation into the beamlet, while causing a shift in the spot placement, did not result in a clinically significant difference in the IDD whereas a shift away from the beamlet’s central axis would due to the increased portion of partial transmission through the trimmer blade.

4.B. Measured and simulated profiles

The incorporation of a DCS prototype into a clinical dose delivery system offers a vehicle to study the influence of trimmer alignment with the beamline with Monte Carlo calculations. Most importantly, the results of these measurements provides a basis to benchmark Monte Carlo calculation methods to accurately simulate complex calculations of radiation transport, such as would be necessary for clinical implementation. Excellent agreement was achieved among all depth dose and lateral profiles. Changes along the trimmed IDD profiles were a result from the lower-energy scatter from the collimators and range shifter that elevated the entrance region along the depth dose profile. The original DCS beamlet model from Gelover et al. predicted this effect and modeled it using an energy- and trimmer position-dependent linear correction function. However, this work offers some of the first experimental characterization of this effect. Additionally, most TPS beamlet models do not currently account for the influence of collimation scatter on a per-beamlet basis. Therefore, this effect will most likely be quantified holistically with the entire dose distribution using Monte Carlo methods, which have been experimentally benchmarked in this work for MCNP6.

Monte Carlo simulations matched the measured dose profiles on- and off-axis; 96% of measured and simulated points matched within a 1%/1 mm global gamma criteria for all measured lateral profiles with the vast majority agreeing above 98%. However, the divergent point source Monte Carlo modeling method used in this work was found to result in simulated beamlets that were slightly larger than measured using the Lynx scintillator, but this difference is less than the voxel resolution of the Lynx. The lateral displacement of the beamlet’s centroid due to collimation was also found to agree between the simulated and measured profiles to within the measurement’s spatial resolution. While only the magnitude of displacements are listed in Table II, similar agreement was also observed for the magnitude of displacement along each axis.

As can be seen in Fig. 11, the majority of the disagreement resided in low-isodose regions that were less than 5% of the peak centroid intensity. While this is within a lower-intensity signal region, the agreement between the Lynx scintillator and film has been shown to diverge slightly at these lower tail regions of proton beamlets¹⁷ and was also observed in this work. Modeling the changes in the spot intensity as a function of trimmer position was in good agreement with the Lynx-measured profiles. Differences that arose were primarily from instances where the collimators were closer to the central axis, which are more sensitive toward changes in intensity than spots further away.²⁶ For example, a 3.6% difference was observed between measured and simulated output when the Y-trimmer was 0.15 cm from the central axis. While this difference could be due in part by deficiencies in the Monte Carlo modeling of the measurement geometry or transport parameters, this difference is more likely due to uncertainty in the actual beam spot position, which can be on the order of a millimeter.²⁷

4.C. Profile sensitivity toward unfocused collimation

Both the benchmarked Monte Carlo model and Lynx-measured profiles indicate that small changes occur to the relative distribution if a beamlet is collimated on-axis vs off-axis. During delivery, all trimmers and spot positions were referenced to isocenter. Since the DCS was aligned parallel to the lateral isocenter plane, the lateral displacement of a trimmer will coincide with its relative position at isocenter. However, the beamlets that are delivered off-axis will be deflected, and their positions at the plane of the trimmers do not necessarily match their position at isocenter resulting in collimation that is not focused toward the beamlet’s central axis. This is similar to the rotational dependence studied earlier in this work. A beamlet deflected 5 cm off axis was found to have similar changes to its lateral profile as a beamlet delivered on axis with a trimmer rotated 1.25°. This effect was found to qualitatively match between simulated and measured profiles as illustrated in Fig. 12.

4.D. Device activation

The dominate interactions that contributed the most to the total induced activity were p → p + n and p → xn reactions within the range shifter and trimmers. Neutron-induced isotopes were only investigated for thermal absorption as the majority of neutrons generated in the device were below 10 MeV. However, only about 0.5% of the total activity was estimated to be a result from the secondary neutrons. Due to the proton-rich characteristic shared by most of the created isotopes, the dominant source of exposure was from the positron annihilation and subsequent 511kėV gamma rays. The creation of ¹¹C from ¹²C was the dominant reaction within the range shifter with a 20.39 min half-life. Several nuclear reactions were also expected to occur from the dominant isotopes within the Ni-200 alloy including ⁵⁸Ni(p,np) ⁵⁷Ni (35.6 h half-life), 60Ni(p,n) ⁶⁰Cu (23.7 min half-life), and ⁶⁰Ni(p,2n) ⁵⁹Cu (81.5 s half-life). The induced activity from the fixed-current IDD measurements resulted in a measured exposure rate of 20 mR/h in front of the device. While certainly elevated, the device was exposed to several days worth of delivered protons. Lateral profile measurements resulted in an exposure rate of only 40 mR/h an hour after completion of the measurements over the course of a few hours.

5. CONCLUSIONS

This work has demonstrated the successful proof-of-concept integration of an experimental DCS prototype into a clinical dose delivery system. Careful modeling of the primary proton beamline and secondary neutron generation was used to benchmark the Monte Carlo transport to experimental measurements. Great agreement was obtained between measured lateral and integral depth dose profiles with a divergent Monte Carlo point source model of the IBA UN beamline and experimental DCS prototype. Retrospectively, this work serves as a benchmark of the computational studies that have relied on using a simplified DCS model toward the development of more sophisticated prototypes. Progressively, the results from this work have shown the sensitivity of the component alignment and influences of spot deflection toward a non-focused collimator. In this capacity, this work also serves as an experimental basis for the continued development of dynamic collimators, which has been the subject of recent focus for the DCS using Monte Carlo methods to design focused collimators.²⁸ Additionally, the activation of positron-emitting nuclides is also non-negligible and should be considered when designing the encapsulation of a clinical system. While the proposed prototype is not a clinically viable device, it has provided an opportunity to benchmark the work that has previously relied solely on Monte Carlo to parameterize asymmetric proton beamlets and investigate the potential challenges that face clinical implementation of a dynamic collimation system.