Mechanical Characterization and Validation of the Dynamic Collimation System Prototype for Proton Radiotherapy

Abstract

Radiation therapy is integral to cancer treatments for more than half of patients. Pencil beam scanning (PBS) proton therapy is the latest radiation therapy technology that uses a beam of protons that are magnetically steered and delivered to the tumor. One of the limiting factors of PBS accuracy is the beam cross-sectional size, similar to how a painter is only as accurate as the size of their brush allows. To address this, collimators can be used to shape the beam along the tumor edge to minimize the dose spread outside of the tumor. Under development is a dynamic collimation system (DCS) that uses two pairs of nickel trimmers that collimate the beam at the tumor periphery, limiting dose from spilling into healthy tissue. Herein, we establish the dosimetric and mechanical acceptance criteria for the DCS based on a functioning prototype and Monte Carlo methods, characterize the mechanical accuracy of the prototype, and validate that the acceptance criteria are met. From Monte Carlo simulations, we found that the trimmers must be positioned within ±0.5 mm and ±1.0 deg for the dose distributions to pass our gamma analysis. We characterized the trimmer positioners at jerk values up to 400 m/s³ and validated their accuracy to 50 μm. We measured and validated the rotational trimmer accuracy to ±0.5 deg with a FARO^® ScanArm. Lastly, we calculated time penalties associated with the DCS and found that the additional time required to treat one field using the DCS varied from 25–52 s.

1 Introduction

Pencil beam scanning (PBS) proton therapy is a technique that uses a magnetically deflected beam to deliver an intensity-modulated dose distribution within a volume [1]. Compared to traditional passive scattering or uniform scanning methods, PBS does not require patient-specific apertures or compensators, which reduces neutron production and total integral dose absorbed by the patient [2–5]. However, for PBS delivered without collimation, the achievable lateral dose conformity depends on the incident beam spot size and the spread of the beam in the patient due to multiple Coulomb scattering [6]. Although achievable beam sizes are shrinking due to improved beam focusing (beam convergence to a smaller spot) systems, the lateral dose fall-off remains one of the limiting factors impacting dose conformity in PBS proton therapy [7]. To achieve clinically acceptable lateral dose conformity for sensitive organs such as the brain, collimating hardware such as a multileaf collimator (MLC) can be used [8,9].

A dynamic collimation system (DCS), shown in Fig. 1, is under development to improve lateral conformity for PBS proton therapy. The DCS consists of two pairs of perpendicular nickel collimators, or trimmers, that move in synchrony with the proton pencil beam to intercept and effectively “trim” the beam at the edge of the target [10]. Each pair of nickel trimmers is responsible for a single axis (X or Y) and is on separate levels. The trimmers independently rotate based on lateral position, effectively providing focused collimation for off-axis pencil beams and ensuring consistent lateral conformity [11]. Treatment planning studies have shown that the DCS can substantially reduce the dose to healthy tissue for brain or head and neck cancers [12–14]. Herein, we determine the dosimetric and mechanical acceptance criteria for the DCS based on a functioning prototype and Monte Carlo methods and measure the mechanical accuracy of the prototype to determine if the acceptance criteria are met. Through this process, the linear positioner motion was characterized, and the kinematics optimized to enable clinical DCS usage with a minimal delivery time penalty.

Fig. 1.

Model of the DCS showing inner to outer layers. The four nickel trimmers (gold) are individually driven by linear positioners (light grey) and each pair (upper and lower) are responsible for a single axis (X or Y). The guide rails (red) help rotate the trimmers to match the beam deflection angle.

2 Materials and Methods

2.1 Sensitivity and Acceptance Criteria.

Highly collimated beamlets are very sensitive to collimator positional or rotational errors [15]. For this reason, the mechanical system must be capable of both high precision and accuracy in a variety of clinical settings. To quantify the level of accuracy required for the DCS, we performed Monte Carlo simulations to determine the sensitivity of dose distributions to the trimmer position and rotation.

2.1.1 Beam Model.

A TOPAS/Geant4-based Monte Carlo model of the Ion Beam Applications (IBA) Proteus Plus at the Miami Cancer Institute (MCI) was created and validated to facilitate the sensitivity studies described in this work [16–18]. The MCI beamline has a nominal energy range of 70 MeV–226.5 MeV with achievable ${\sigma }_{\mathit{air}}$, which represents the Gaussian sigma of the lateral intensity distribution in air, ranging from 7–3 mm at isocenter, respectively. Because each spot does not have a consistent fluence (protons per unit area) across it, we characterize the intensity of the lateral profile using a Gaussian and the sigma of the Gaussian is used to represent the spot size or lateral width. A 4-cm thick polyethylene range shifter, which attenuates protons to allow treatment of energies below 70 MeV, is included in the model. The initial energy and energy spread were modeled using Gaussian functions, and all simulations were validated to >99% passing rate for a gamma index of 1%/1 mm using measurements taken at MCI [19]. The gamma index is a popular metric that combines percent dose difference and distance-to-agreement into a single value to determine similarity between dose distributions [20]. For our simulations, we chose thresholds of 1% dose difference and 1 mm distance to agreement; both of which are considered strict. For analyses of treatment plans in the clinic, looser gamma thresholds of 3% and 2 mm are generally used.

2.1.2 Sensitivity Simulations.

A sensitivity study based on Monte Carlo simulations was used to define the mechanical accuracy requirements for the DCS. The simulation geometry, shown in Fig. 2, included a single trimmer placed with the face of the trimmer 1.5 mm away from the beam central axis. The three other trimmers were removed from the field. Energy of 156.5 MeV with an energy spread of 0.967 MeV/0.62% was selected for these simulations. At this energy and with the inclusion of the range shifter, the protons have a range in the water of 11.6 cm, a reasonable therapeutic depth for brain cancer and head and neck cancer target volumes. After the ideal geometry was simulated, comparative simulations were performed by independently introducing positional and angular errors ranging from 0–1 mm and 0–1.0 deg.

Fig. 2.

Basic diagram of the TOPAS setup showing the range shifter, trimmer, and water phantom

Five million protons were simulated for each setup, enough to keep the standard deviation of the dose below 0.5% of the maximum.

2.2 Control System Design.

The accuracy of radiation dose delivery using the DCS depends on the precise and automatic positioning of multiple trimmers during treatment delivery. Each trimmer is positioned by a Parker T1S linear positioner and controlled by an EtherCAT (Ethernet for Control Automation Technology) motion control system from ACS Motion Control (Yokneam Illit, Israel). The controller ensures safe DCS operation by limiting the drive current and protecting against overloads, faults, and trimmer collisions. It also provides the capability to collect data and implement custom control logic.

The control system, shown in Fig. 3, consists of a motor driver module, an analog input module to receive setpoints for the trimmer positions for each spot delivery, an analog output module to provide feedback about the current trimmer positions, and a controller to interface between the modules using ACS Motion Control software and the clinical proton therapy system. The controller has a cycle time of 0.2 ms (5 kHz).

Fig. 3.

A schematic of the DCS control system. Connections colored by their type: red—power, black—digital data, green—analog out, and blue—analog in. The IBA Scanning Controller refers to the component on the proton therapy delivery system that controls the treatment delivery and will be responsible for providing commands to the DCS.

The linear positioners are equipped with encoders to track motion and provide feedback with a resolution of 50 μm. Since the encoders measure relative position, calibration is necessary for accurate absolute positioning. To achieve this, the DCS has built-in calibration blocks at known positions for the encoders to home against when the system is powered on. At startup, each motor slowly translates the trimmer across the field of view until the trimmer physically contacts the calibration block. When the control system detects the contact, the system records that position to define the mechanical coordinate system for the trimmers.

Each calibration block is also fitted with a miniature pneumatic shock absorber that acts as a safety measure for impact absorption for the situation in which the DCS linear positioners are oriented vertically and a motor loses power or is disabled. However, the primary provision available if the linear positioners lose power while in a vertical orientation is dynamic braking. If the axis is disabled and the feedback velocity is less than the braking velocity, the brake is applied dynamically by shorting the drive and motor windings together. The energy from the motor winding is then routed back into the motor to counterthe motion.

2.3 Mechanical Characterization

2.3.1 System Tuning.

DCS motor tuning ensures stability to physical disturbances and allows the system to counteract the inertia of each axis. Without tuning, motor movement may be under- or over-damped. Tuning the system is an automated process handled by the ACS control software that cycles the motors through a range of accelerations to determine the optimal acceleration curve. The optimal acceleration curve is one that maximizes velocity without inducing any resonance into the system that could compromise stability. Because tuning is a function of the overall mechanical properties of the DCS and the system it will be mounted to (rigid proton gantry), a test bench was constructed to replicate a rigid environment for tuning the DCS. The test bench consisted of an exoskeleton built using 1.5” aluminum modular 80/20^® extrusions (Columbia City, IN). This exoskeleton, with the DCS mounted within, was then attached to a milling machine to replicate the mass of a gantry and the automated tuning program was executed.

2.3.2 Characterizing Jerk Values.

To determine optimal operating speeds, the DCS was tested with preprogramed sequences using various stroke lengths and motor jerk values, where jerk is defined as the time derivative of acceleration. The stroke lengths were 1, 2, 5, 10, 20, 40, 80, 100 and 120 mm with jerk values of 50, 100, 200, 300 m/s³. Within the ACS controller software, while velocity and acceleration are adjustable parameters, the jerk value ultimately limited trimmer velocity and acceleration due to the short stroke lengths. From the ACS software, encoder position and timing data can be extracted to evaluate the movement behavior and calculate stroke times for given jerk values.

2.3.3 Calculated Treatment Times Based on Test Plan.

Minimizing treatment time is important for patient comfort, to minimize intrafraction patient motion, and ensuring efficient clinical operations. Across all tumor sites, an average PBS proton therapy treatment plan will take about one minute per field to deliver, and a design goal of the DCS is to increase this delivery time by no more than one additional minute per field.

An example treatment was simulated for a braincase, and the associated trimmer positions for each beam spot were imported into the DCS control system [21]. The overall time required for the trimmers to visit each required trimmer position, along with a 0.3 s dwell time after reaching the position, was recorded for jerk values ranging from 50 to 400 m/s³ to quantify the dependency of delivery time on jerk setting.

2.4 Positional Accuracy

2.4.1 Positional Accuracy/Encoder Validation.

Stroke lengths of 1, 2, 5, 10, 20, and 40 mm lengths were performed and accuracy was measured using a three-dimensional (3D) FARO^® Edge ScanArm (Lake Mary, FL), which is accurate to ±25 μm. Each positional accuracy measurement was calculated using an average of four separate points to minimize setup error.

2.4.2 Operation Stability.

To determine if the encoders remain spatially accurate after initialization, trimmer positions at fixed setpoints were recorded, the DCS motors were then instructed to move for 30 min, and the trimmer positions at the same encoder locations were remeasured using the FARO^® ScanArm. This is expected to be a representative stress test given the timeframe of 1–2 min of continuous use per patient.

2.4.3 Effects of Rotational Orientation.

The DCS has been designed to attach to a delivery system capable of rotating 360 deg around the patient. This means that the DCS must be capable of accurate operation in different rotational positions. To mimic this setup, the DCS was mounted inside the same aluminum exoskeleton from previous tests which allowed the DCS to hang from its mounting points as if it were attached to a proton therapy nozzle. When the exoskeleton is rotated, the DCS will be under the same stresses as if it were mounted to the nozzle.

To investigate the effect gravity has on the DCS for different orientations, a preprogrammed treatment plan test sequence was executed for the same range of jerk values as performed in the previous section. The two orientations tested were 90 deg (sideways) and 180 deg (upside down). Although in practice the DCS may be operated in an arbitrary position with respect to gravity, we found it sufficient to test the orientations where the effects of gravity on the trimmer blades are at a maximum or a minimum.

2.5 Angle Accuracy

2.5.1 Mechanical Tolerance of Angle Mechanism.

The trimmer focusing mechanism uses an attached lever arm that follows a slotted guide rail, Fig. 10, providing trimmer rotations with linear position [11]. The slotted guide rail was milled with a ±100 μm tolerance to avoid mechanical seizing with the pin that rides in the rail. This tolerance in the guide rail results in the potential for a small variation in angle for a given position. To characterize this variation, the trimmer was manually rotated to its extremes while in a stationary position, and the angles were measured using the 3D FARO^® Edge ScanArm.

Fig. 10.

The bearing (green) follows the adjustable slotted rail (red) that raises or lowers the lever arm, causing rotation of the trimmer (gold)

2.5.2 Angle Accuracy Validation.

The trimmer angles were specifically tuned for the source-to-collimator distances (SCD) of the IBA Proteus Plus at the Miami Cancer Institute. Because the bending magnets are sequential in the beamline, the point of beam deflection is in different locations for each axis and a matching pair of trimmers. This means that the angle change as a function of the translation will be different for each axis. For the top pair of trimmers, a rotation of 0.35 deg per centimeter of translation is expected, whereas the bottom pair rotate at 0.30 deg per centimeter. Using a 3D FARO^® Edge ScanArm, we measured the trimmer angles at −50, 0, and +50 mm positions on their translation axes. Using the previously stated accuracy of the 3D FARO^® Edge ScanArm (±25 μm), we estimate the accuracy of the angles measured to be ±0.08 deg.

3 Results

3.1 Sensitivity and Acceptance Criteria.

Monte Carlo results show that the trimmers must be positioned within 0.5 mm to maintain a 100% passing rate, every point having a value < 1, for gamma criteria of 1 mm and 1%, as seen on the left of Fig. 4. The right side of Fig. 4 shows how a 1 mm error effects the lateral dose distribution; and as seen in the red, the gamma analysis fails. As for the angular threshold, we found that the trimmers must be within ± 1.0 deg to achieve 100% gamma index passing rate. As a result, we selected ± 0.5 mm and ± 1.0 deg as our target design criteria.

Fig. 4.

Lateral dose distribution from collimated spots with induced 0.5 mm positional errors. The reference profiles are in black, the ± 0.5 mm errors are in blue and gamma index (1%/1 mm) is in red.

3.2 Mechanical Characterization

3.2.1 System Tuning.

Prior to mechanically tuning the DCS, the system exhibited an underdamped response which resulted in overshoots as much as 0.5 mm. After tuning, the motor movement was smooth and critically damped for all jerk values. Encoder data from before and after tuning can be seen in Figs. 5 and 6, respectively.

Fig. 5.

Encoder data showing trimmer position versus time before the DCS was tuned. On the right is a zoomed in frame showing detail of trimmer movement. Black is a jerk value of 50  $\mathrm{m}/{\mathrm{s}}^3$ and red is 100  $\mathrm{m}/{\mathrm{s}}^3$.

Fig. 6.

Encoder data showing trimmer position versus time after the DCS is tuned. On the right is a zoomed in frame showing detail of trimmer movement. Black is a jerk value of 50  $\mathrm{m}/{\mathrm{s}}^3$ and red is 100  $\mathrm{m}/{\mathrm{s}}^3$.

3.2.2 Characterization of Jerk Values & Treatment Times.

After successful tuning, the DCS was capable of operating at jerk values up to 400  $\mathrm{m}/{\mathrm{s}}^3$ without vibration or dampening issues. However, we did not execute every test at 400  $\mathrm{m}/{\mathrm{s}}^3$ because it is faster than what is necessary clinically and results in more audible noise. From the encoder data and a sample treatment plan, time penalties associated with DCS treatments were calculated for each jerk value. The sample/uncollimated treatment plan was optimized for the DCS to minimize treatment time, and we calculated the plan would take 73 s to complete, assuming a 0.3 s dwell time at each required proton beam spot location [21]. As seen in Fig. 7, the DCS treatment time penalty ranged from 25–52 s, depending on jerk value.

Fig. 7.

Motor jerk value versus treatment time. Percentages shown are the increase over the base treatment time of 1973s.

3.3 Positional Accuracy

3.3.1 Positional Accuracy/Encoder Validation.

The encoders for each axis were validated to within their ±50 μm resolution. For all the axes and stroke lengths, there was only a single measurement that was 50 μm from the expected value. Figure 8 shows the average error and corresponding standard deviation for each axis.

Fig. 8.

The mean and standard deviation of each axis' accuracy. Individual points also plotted.

3.3.2 Operational Stability.

The DCS was operated continuously for 30 min and the average encoder drift was measured to be 25 micron for all four axes. The largest drift was measured to be 42 microns, as shown in Table 1.

Table 1.

Motor drift after 30 min of continuous operation

	Axis 1	Axis 2	Axis 3	Axis 4	Avg.
Drift (mm)	0.006	0.019	0.042	0.034	0.025

3.3.3 Effects of Rotational Orientation.

After running the preprogrammed test sequence using jerk values up to 300  $\mathrm{m}/{\mathrm{s}}^3$, the encoder data, shown in Fig. 9, indicated that the motors, drivers, and controllers are capable of maintaining the speed and accuracy while under different gravitational forces.

Fig. 9.

Motor position versus time for three different orientations run at 300 m/s³ jerk. Data is from onboard encoders.

3.4 Angle Accuracy

3.4.1 Mechanical Tolerance of Angle Mechanism.

The “slop” in the angle due to the mechanical tolerances of the slotted rail mechanism was measured to be ±0.25 deg. While operating, the pin tends to track along a singular side of the rail causing the operating slop to be effectively half of what was measured for any given orientation. For example, in Fig. 10, as the trimmer is translated right, the bearing (green) will track along the bottom of the guide rail (red).

3.4.2 Angle Accuracy Validation.

The trimmers were successfully adjusted to within ±0.25 deg of the desired angles for all four axes, with an average error of 0.114 deg. Figure 10 shows the set screws (black) on the top and bottom of the guide rail (red) used to adjust the guide rail and trimmer angles. The largest error measured was 0.204 deg at a single location for a single trimmer. Figure 11 is a plot of the trimmer angles versus position with the desired angle in red.

Fig. 11.

Trimmer angle versus position is shown for each axis. The top pair of trimmers is on the left and the bottom pair is on the right. The ideal angle is shown in red.

4 Discussion

Monte Carlo methods were used to investigate and quantify the sensitivity of the dose distributions with respect to the positional and angular errors of the trimmers. Based on simulation results, we selected a relatively conservative threshold of ±0.5 mm and ±1.0 deg as our acceptance criteria. These are chosen because when errors of this magnitude are induced, the dose distributions remain very similar and still pass a gamma of 1%/1 mm for 100% of the points. These accuracy thresholds ensure that the dose distributions will not be substantially impacted by collimator inaccuracies. We successfully validated the true positional and angular accuracy of each trimmer to below our thresholds.

Proper tuning of a mechanical system is vital for smooth and reliable operation. Once tuned, the DCS is capable of very fast movements and changes of direction, using a jerk value of up to 400 m/s³. From a 73 s sample treatment time, the DCS incurs a time penalty of 25–52 s or 34.2%–71.2%, depending on the jerk value.

We expected the Parker T1S linear positioners and built-in encoder to meet the accuracy requirements, and the FARO arm measurements confirmed that expectation. The encoders did not drift over a period of 30 min of continuous operation and the device orientation had no measurable effects on the positional accuracy or stroke time.

It is important to note that positional errors have a much larger effect on the beam spots than angular errors. For example, a 0.5 mm positional error will induce a ∼0.5 mm shift in the beam spot position (Fig. 4), but a ±1.0 deg angular error will not induce an appreciable shift in the high dose gradient region of the beam spot [22]. The angular error instead primarily affects the low dose tail along the collimated edge, shown in Fig. 12 and discussed in Geoghegan et al. [11]. This is a favorable property of the DCS because achieving the desired accuracy of the trimmer angles is more difficult than for trimmer positions, due to the required operational tolerances to allow slippage in the guide rails compared to feedback from the built-in positional encoders.

Fig. 12.

Lateral profiles of a trimmed beam spot with ±1 deg errors induced. Arrow pointing to low dose tail.

Lastly, the rotation mechanism is designed so that, barring any mechanical failure, the trimmers cannot be more than ±0.25 deg at a single position. This ensures that no matter the rotational orientation of the DCS and the resulting change in gravitational force, the trimmer angles will be within specification.

5 Conclusion

Using Monte Carlo methods, we first determined the trimmer positional and angular thresholds (0.5 mm and 1.0 deg) that must be met in order to ensure accurate dose delivery while using the DCS. Then, using a variety of tools and techniques, we successfully characterized the kinematics and validated the positional and angular accuracy of the DCS to be within our desired thresholds. We have shown that the DCS is capable of operating at jerk values up to 400 m/s³ for different orientations (0 deg, 90 deg, and 180 deg) and maintaining accuracy within ± 50 microns. Lastly, we demonstrated that we could operate the DCS at a high enough jerk value to maintain a treatment time penalty of under one minute per field.

Conflict of Interest

Hyer, Flynn, and Hill are co-inventors on a patent that has been licensed to Ion Beam Applications (IBA).

Funding Data

• National Cancer Institute (Grant No. R37CA226518; Funder ID: 10.13039/100000054).

Nomenclature

DCS =

dynamic collimation system

MLC =

multileaf collimator

PBS =

pencil beam scanning