A sparse orthogonal collimator for small animal intensity-modulated radiation therapy. Part II: hardware development and commissioning

Abstract

Purpose:

A dose-modulation device for small animal radiotherapy is required to use clinically analogous treatment techniques, which will likely increase the translatability of preclinical research results. Because the clinically used multileaf collimator (MLC) is impractical for miniaturization, we have developed a simpler, better-suited sparse orthogonal collimator (SOC) for delivering small animal intensity-modulated radiation therapy (IMRT) using a rectangular aperture optimization (RAO) treatment planning system.

Methods:

The SOC system was modeled in computer-aided design software and fabricated with machined tungsten leaves and three-dimensional (3D) printed leaf housing. A graphical user interface was developed for controlling and calibrating the SOC leaves, which are driven by Arduino-controlled stepper motors. A Winston-Lutz test was performed to assess mechanical alignment, and abutting field and grid dose patterns were created to analyze intra- and intercalibration leaf positioning error. Leaf transmission and penumbra were measured over the full range of gantry angles and leaf positions, respectively. Three SOC test plans were delivered, and film measurements were compared to the intended dose distributions. The differences in maximum, mean, and minimum, as well as pixelwise absolute dose differences, were compared for each structure, and a gamma analysis was performed for the target structures using criteria of 4% dose difference and 0.3 mm distance to agreement.

Results:

The Winston-Lutz test revealed maximum directional offsets between the SOC and primary collimator axes of 0.53 mm at 0° and 0.68 mm over the full 360°. Upper and lower abutting field patterns had maximum dose deviations of 18.8 ± 3.1% and 15.5 ± 2.9%, respectively, and grid patterns showed intra- and intercalibration repeatability of 93% and 91%, respectively. Extremely low midleaf (0.15 ± 0.05%) and interleaf (0.27 ± 0.22%) transmission was measured, with no significant rotational variation. The average penumbra was ~0.8 mm for all leaves at field center, with a range of 0.17 mm for all leaf positions. A highly concave test plan was delivered with a ~ 95% gamma analysis pass rate, and a realistic mouse phantom liver irradiation plan achieved a pass rate of ~98%. A highly complex dose distribution was also created with 551 SOC apertures averaging 2.4 mm in size.

Conclusions:

A sparse orthogonal collimator was developed and commissioned, with promising preliminary dosimetry results. The SOC design, with its limited moving components and high dose-modulation resolution, is ideal for delivering high-quality small animal IMRT with our RAO-based treatment planning system.

1. INTRODUCTION

While modern radiotherapy has made strides in early-stage lung, liver, metastatic, and other radioresistant cancers, there is a need to further improve the treatment efficacy of radiotherapy for challenging diseases including glioblastoma multiforme (GBM),¹ locally advanced lung,^2,3 liver,^4-6 and pancreatic cancers.^7-9 Future improvements in the efficacy of radiotherapy may rely on innovatively combined immune and radiotherapy,^10-14 x ray dose triggered drug release,^15-17 spatially modulated radiotherapy targeting tumor cell heterogeneity and niches,^18,19 temporally modulated radiotherapy to disrupt tumor stem cell proliferation,²⁰ or ultra-high-dose radiotherapy.^21-26 However, because of the huge variable space for these experimental therapy techniques and the intrinsic risk associated with interventions involving radiation and/or molecular agents, sound animal experiments must be performed before advancing to human trials. To produce translatable preclinical results, animal irradiation techniques need to closely mimic human treatment. For example, it has been shown that radiation-induced lymphopenia, which diminishes the efficacy of PD-1 immune checkpoint inhibitors,²⁷ is correlated with the irradiated normal tissue volume.^28,29 Because of the large gap in dose conformity between preclinical and clinical radiotherapy, an animal lung tumor model to test the efficacy of combined immune and radiotherapy with whole or hemi-thoracic treatment would be highly inconsistent in immune response with a human trial using the conformal SBRT technique.

In addition to oncological applications, radiation treatment to morphologically complex normal mouse organs, including the brain,³⁰ lung,³¹ liver,^32,33 and bone marrow,³⁴ is commonly performed to study the effects of radiation. Without IMRT, excessive high dose spillage to the surrounding unintended tissues would unavoidably contaminate the causal relationship between irradiation of the intended target and its effect. For example, in whole-brain treatment, high dose spillage to the oral cavity will cause mucositis that adversely impacts animal feeding, and subsequently survival, and behavior that may be misinterpreted as brain irradiation effects.

To deliver analogous treatment to small animals, the development of a small animal IMRT platform is necessary. Such a platform requires two developments, the first being a planning approach based on inverse fluence map optimization, such as the rectangular aperture optimization (RAO)-based treatment planning system detailed in Part I. The second is a dose-modulation device suitable for the small animal scale.

The majority of small animal radiotherapy systems have been limited to stationary collimators, with a few exceptions. At Stanford University, a variable hexagonal aperture collimator was developed to achieve beam diameters up to 102 mm.³⁵ The University of Western Ontario developed motorized orthogonal jaws for their micro-CT/RT system, with characterization focused on 2 to 30 mm field sizes.³⁶ A binary micro-MLC (bmMLC) has also been developed by the University of Wisconsin-Madison.³⁷ Unlike the dynamic MLC with a continuous range of leaf positions, the bmMLC leaves are either open or closed, similar to the collimator design for the tomotherapy system.³⁸ However, this system is limited by excessive leakage (5.4% interleaf leakage, 1.7% average transmission) and long delivery times. While all of these devices enable more conformal treatment delivery, none are optimal for small animal IMRT.

More recently, a motorized variable collimator (MVC) has become commercially available for the small animal radiation research platform (SARRP) (Xstrahl Inc., Suwanee, GA), which uses two sets of focused, orthogonal tungsten jaws. Very preliminary efforts using the MVC for jaws-only small animal IMRT have been presented from the Institute of Cancer Research, London.³⁹ This approach generates IMRT fluence maps first and then converts them to rectangular apertures. This process is known to degrade the dosimetry, which worsens when the apertures are limited in shape and number. Furthermore, the MVC is fundamentally limited by its inability to create asymmetric apertures, requiring the use of couch movements to deliver off-axis fields. This excessive couch translation introduces significant potential for motion errors and increased delivery times, which are already lengthy because only one aperture can be delivered at a time.

Clinical dose-modulation devices have been developed to exploit the unique advantages of bidirectional collimation while also providing the flexibility of an MLC. In studies on the commissioning of a dual-layer micro-MLC for clinical radiotherapy systems, the additional layer of leaves significantly reduced the leaf-end transmission,^40,41 which can be as high as 30% for a typical single-layer MLC.⁴² The ability to collimate fields in both directions with the second orthogonal layer MLC layer also enables higher resolution than a standard single-layer MLC, for which the resolution perpendicular to leaf motion is determined by the physical leaf width.

The sparse orthogonal collimator (SOC), with four pairs of double-focused orthogonal leaves, maintains the simplicity of a jaws-only system while achieving flexibility closer to that of a dual-layer MLC, as well as increased delivery efficiency. This makes the SOC ideal for small animal dose modulation when used in combination with the RAO-based treatment planning system detailed in Part I. In this study, the complete design and fabrication of the SOC hardware is presented, as well as the software development and full system commissioning and testing.

2. MATERIALS AND METHODS

2.A. SOC design

In order to deliver rectangular apertures, theoretically only two pairs of orthogonal leaves are required (essentially “jaws-only” IMRT). However, increasing the number of leaves increases the delivery efficiency, as it enables the dose to certain regions of the fluence map to be delivered in parallel, as previously mentioned. Increasing this to more than two leaves per bank greatly increases the complexity of the system while providing only minor improvements in efficiency, as previously shown.⁴³ Therefore, the preclinical SOC designed in this study features two leaves per bank.

This orthogonal design is ideal for a miniaturized small animal system as the deliverable aperture resolution is dependent on the leaf motion step size rather than the leaf width. However, the simplicity of the SOC design requires that the two sets of orthogonal collimator leaves have slightly different source-to-leaf distances, which could cause the delivered fluence rectangles to have an asymmetric penumbra. This is mitigated with a double-focused leaf design, where the leaves have curved geometry and move along an arc that matches the beam divergence. While the general SOC geometry is feasible for use with a variety of irradiators, the design detailed in this study was tailored to the X-RAD SmART small animal image-guided irradiation system at UCLA. Computer-aided design (CAD) models for all SOC system components were created using Autodesk Inventor (Autodesk, Inc., San Rafael, CA).

Designs for the double-focused SOC leaves are shown in Fig. 1. The radii of curvature were chosen for a source-to-leaf distance (to the top of the upper leaf) of 14.94 cm, with 3mm-thick leaves and a gap of 0.5 mm between upper and lower leaf sets to avoid collision. The leaves are 3 mm thick to achieve >99.5% beam attenuation, based on preliminary transmission measurements with the same tungsten alloy (95% W, 3.5% Ni, 1.5% Fe; density of 18 g/cm³). The depth of the tongue and groove is 0.2 mm in order to keep the leaves interlocked without significantly affecting the penumbra. The heights of the tongue and groove are 1.4 and 1.6 mm, respectively. These dimensions are ideal firstly because a 0.1 mm tolerance is necessary on each side of the tongue for smooth movement. Also, this means that both leaves have a total of 1.4 mm of tungsten extending the 0.2 mm of the tongue and groove joint and will therefore have the same x ray attenuation, preventing uneven SOC aperture penumbras. Each leaf also has a 1.35-mm-thick, 1.45-mm-wide tongue that slides along rails in the SOC housing for added leaf stability. These tongues are along the top edge for upper leaves and bottom edge for lower leaves. Each upper and lower leaf pair also has a mirroring pair for the opposing leaf bank, with the leaf connector tab on the opposite end, enabling interdigitation of opposing tongue and groove joints. Therefore, the geometry for each of the eight leaves is unique.

Fig. 1.

Leaf designs for the sparse orthogonal collimator, with dimensions shown in millimeters. Upper and lower leaf sets feature different radii of curvature and different geometry for the outer stabilizing tongues (along the top/bottom for upper/lower leaves, respectively). Each tongue and groove pair also has a mirroring pair with the tab on the opposite end.

The CAD model for the rest of the SOC system is shown in Fig. 2. A custom rod with clevis and pin attaches the tab on each leaf to a stepper motor, which screws into a support structure in the housing. The outer leaf tongues slide along upper or lower rail systems for stability. A column of small (4 mm) tungsten cubes at the intersection of each orthogonal leaf bank shields the outside corners of the primary collimator field (4 × 4 cm at isocenter), which is larger than the SOC field (2.6 × 2.6 cm). The top of the housing features a detachable cable guide for the motor wires, which attach to a common socket at the back of the SOC. The housing top also has an adapter for mounting the system to the primary SmART collimator. The adapter is designed for a tight fit onto the collimator (which unfortunately has no other features to use for mechanical attachment) with a screw clamp for extra support. Figure 3 shows the whole SOC system mounted onto the SmART gantry.

Fig. 2.

Computer-aided design model of the sparse orthogonal collimator (SOC). (a) Leaves connected to motor shafts with three-dimensional printed rods and pins. (b) Stepper motor pair screwed into housing. (c) Outer leaf tongues slide along rails in housing. Shown in each corner of the SOC field are tungsten inserts for shielding the corners of the larger 4 x 4 cm primary collimator field. (d) All four orthogonal leaf banks. (e) Covers slide into the housing over each motor bank. (f) Main housing top attached, with mount for square primary collimator. (g) Complete SOC system with adapter to the primary 4 × 4 cm square SmART collimator. The collimator slides tightly into the adapter with a screw clamp for added support. The C-shaped attachment around the adapter guides the motor cables to the main socket on the back of the SOC.

Fig. 3.

The sparse orthogonal collimator (SOC) mounted onto the gantry of the SmART system via the primary 4 cm fixed collimator.

2.B. SOC fabrication

The four pairs of leaves for the SOC were machined out of a class 3 tungsten alloy (95% W, 3.5% Ni, 1.5% Fe; density 18 g/cm³) by a machine shop (Fig. 4). While tungsten is considerably more difficult to machine than a softer metal such as copper, its high density is necessary for achieving the desired attenuation with leaves thin enough for the small scale of the SOC.

Fig. 4.

Machining of the prototype tungsten alloy leaves for the sparse orthogonal collimator.

The main housing for the SOC, shown in Fig. 5, was fabricated out of plastic with stereolithography three-dimensional (3D) printing (Formlabs Form 2, White Resin). The primary collimator adapter, which attaches the SOC to the 4 cm square collimator of the SmART system, was also 3D printed. Tungsten cubes (4 mm) were inserted into the housing between perpendicular leaf banks for shielding, since the primary collimator field is larger than the open SOC field.

Fig. 5.

Three-dimensional (3D) printed sparse orthogonal collimator (SOC) housing: main SOC leaf housing (upper left); outer leaf tongues slide along housing rails (upper middle); leaves attached to motors with 3D printed rods and pins (upper right); housing with all 8 SOC leaves (lower left); housing top with opening for 4 x 4 cm primary collimator and cable guide attachment (lower middle); and housing for Arduino board and motor driver boards (lower right).

The SOC leaves are driven by captive stepper motor linear actuators (Haydon Kerk Motion Solutions, Size 8) which have a maximum speed of over 4 cm/s and resolution of 0.04 mm/step. Rotary encoders are installed for precise leaf position verification with a resolution of <0.02 mm. The motors are wired to stepper motor driver boards (EasyDriver V4.5), which are controlled with an Arduino Mega 2560 microcontroller. These boards enable microstepping of the motors to achieve even finer motor control of 0.005 mm per eighth step.

2.C. SOC control software

A graphical user interface (GUI) has been developed in Python for controlling the SOC leaf motion (Fig. 6). The leaves can be moved by either manually dragging the leaf images in the GUI or by loading an aperture sequence from a JavaScript Object Notation (JSON) file. The aperture sequences can also be created or edited within the GUI and saved as a JSON. Leaf motion commands are then sent over a serial connection to the Mega 2560 board, which is loaded with an Arduino sketch. This sketch evaluates the new leaf positions for any potential collisions (and sends an error if necessary) and then moves each motor the necessary number of steps, using the encoder feedback for verification. The controller proceeds with the treatment once it receives confirmation of successful leaf repositioning. There is also a leaf calibration feature, which allows the user to set a new “zero” position for the leaves.

Fig. 6.

Graphical user interface for sparse orthogonal collimator control, with an aperture sequence loaded from a JavaScript Object Notation (JSON) file. Apertures can be edited and saved using the Sequence Item Editor shown. Leaves can also be controlled by dragging the gray “leaves” on the left. The “Calibrate Leaflet Positions” feature can also be used to set a new leaf “zero” position.

These aperture sequence files are created in MATLAB following fluence map optimization with RAO and plan postprocessing, as described in Part I of this report. The apertures are first shifted to the center of the SOC field, if necessary. The aperture boundaries are converted to a set of eight leaf positions, which are then written to a JSON file for loading into the SOC controller GUI. The couch angle (optional using a 3D printed rotating couch mount, shown in Fig. 9), gantry angle, and dwell time are also included, along with an option for manual or automatic sequence progression.

Fig. 9.

(Left) Mouse phantom modeled from mouse computed tomography data and three-dimensional (3D) printed with a flexible, tissue-equivalent material and a midcoronal split for film measurement. Phantom is shown on the previously mentioned rotating couch mount. (Right) 3D printed block phantom for axial dose measurements.

2.D. SOC installation

Prior to initial installation and commissioning of the SOC, the SmART calibration software was used to verify the source-axis distance (SAD), magnification factor, isocenter location, and center of the detector panel. A Winston-Lutz map of the primary 4 cm square collimator was also created with the system calibration software to factor into the overall system alignment measurements. For this test, PXI’s standard 3 mm calibration BB phantom is first moved to the treatment isocenter. Using the projection on the flat panel detector, the calibration software then measures the offset between the center of the collimator field and the treatment isocenter over a full gantry rotation.

The SOC is installed on the head of the SmART by attaching it to the primary collimator with a 3D printed sleeve and clamp (Fig. 7). Cabling from the electronics (driver boards and microcontroller) to the SOC is fed through a side hole in the cabinet, with shielded cable used inside the cabinet for noise suppression. The leaf positions are then calibrated using the fluoroscopy mode of the SmART calibration software. The leaves are extended individually to measure the position of the x- and y-axis of the SOC, determined by the inner edges of the upper and lower leaves, respectively (Fig. 8, upper). The leaves are then moved the appropriate distance from each axis for an open field, based on the desired field size at isocenter (maximum of 26 × 26 mm, defined by the FWHM) and the measured magnification factor of the system to the flat panel detector. Using the calibration module of the GUI, these leaf positions are saved as the new “zero” positions for the motor encoders (Fig. 8, lower).

Fig. 7.

Sparse orthogonal collimator installed on SmART system, attached to the 4 x 4 square primary collimator.

Fig. 8.

Sparse orthogonal collimator (SOC) leaf position calibration procedure using the fluoroscopy mode of the SmART system calibration software. (Upper) SOC axes are determined from the inner leaf edges. (Lower) Open leaf positions are set to the desired distance from each axis, and the new calibration is saved with the SOC GUI.

To verify that the leaves are properly aligned to each other and at the correct field size, a simple grid pattern is then delivered to film and the resulting dose distribution is analyzed. A custom couch-mounted 3D printed film holder is used with standard-sized film squares for reproducible alignment.

2.E. SOC commissioning measurements

After initial installation, the mechanical accuracy and reproducibility of the SOC were evaluated. The AAPM Task Group 50 report, “Basic Applications of Multileaf Collimators,”⁴⁴ includes a section for MLC Acceptance Testing, Commissioning, and Safety Assessment describing methods for performing many of the necessary QA tests and measurements for a new MLC. Although some procedures could not be directly translated due to hardware limitations (e.g., no light field or collimator rotation), a series of adapted tests were performed. All film measurements were performed with Gafchromic EBT3 film (lot #10161801) (Ashland Inc., Covington, KY), which was scanned on an Epson 10000XL scanner (96 dpi resolution, 48-bit RGB) and analyzed with Ashland’s FilmQA Pro software using a lot-specific red channel calibration curve created according to the manufacturer-recommended protocol.⁴⁵ The calibration film (11 dose levels, 0–20 Gy) was irradiated with a 4 × 4 cm field on the surface of a 30 × 30 × 5 cm solid water slab using the AAPM Task Group 61 protocol for in-air dose calibration.⁴⁶ The system output was measured using a Farmer® 30010 Ionization Chamber with a 250 kV calibration.

First, the mechanical axes alignment and rotational stability of the SOC were verified. Alignment of the SOC with the gantry axis was measured by exposing the same film, placed at the system isocenter, with both the open SOC field and the 40 mm square primary collimator alone and measuring the distance between field center points. The x- and y-axis for each collimator were defined by the center of the 50% isodose distribution. This was repeated at 1 cm below isocenter to reveal any angular misalignment. The rotational stability and effects of the extra SOC weight on the gantry alignment were then evaluated by repeating the Winston-Lutz test with the SOC installed on the primary collimator. A spoke shot pattern was also created for visual verification by exposing film perpendicular to the rotational axis with a 1 mm slit field at 45° increments.

To measure the leaf positioning accuracy, a series of abutting rectangular fields (26 × 4 mm) were delivered to film, and the dose at the matchlines between fields was analyzed. This was repeated for both upper and lower leaf sets (i.e., horizontal and vertical). A grid pattern was then delivered to verify the alignment between the upper and lower leaf sets, and a pixelwise comparison was performed between the 50% isodose distribution and the intended pattern. This grid pattern was delivered several times in a row to assess the leaf position repeatability by measuring the distance between 50% isodose lines. Grids were also compared between different instances of SOC installation and calibration to assess the calibration accuracy.

The transmission through the center of each leaf (midleaf), between each tongue and groove joint (interleaf), and between opposing leaf faces closed at the field midline (closed leaf) was also measured using film. Measurements were performed for both upper and lower leaf tiers, with the four leaves closed at the midline of the field. This was repeated at gantry angles of 90°, 180°, and 270° to identify any changes in leakage from potential leaf shifts during rotation. Transmission was calculated as the maximum leakage dose divided by the maximum open field dose for the same exposure time. The penumbra, calculated as the distance from 80 to 20% of the maximum dose, was determined for upper and lower leaf faces and edges (tongue and groove) by exposing film with each leaf extended individually to the midline. Because the leaves are double-focused, the leaf face penumbra should not change significantly with leaf position. To verify this, the penumbra was also measured for central square apertures with lengths of 26 (open field), 20, 10, and 5 mm at depths of 2 and 20 mm.

2.F. Dosimetric testing

Three different test plans, either analogous (a highly concave target) or identical [the mouse whole liver and complex two-dimensional (2D) “Audrey” plans] to those evaluated in Part I, were delivered for end-to-end testing of the SOC treatment planning and delivery process (Table I). All plans were delivered with a 225 kVp, 20 mA x ray beam with 0.32 mm added copper filtration.

Table I.

Sparse orthogonal collimator (SOC) test plan parameters.

Test plan	# of beams	Target dose levels	Apertures per beam (average)	Average aperture size (mm)	Delivery time (min)	Film measurement setup
C-shaped target	8	1 (10 Gy)	76.6	3.5 ± 1.4	~14	Cube phantom (midaxial plane)
Mouse liver	5	1 (10 Gy)	101.6	1.8 ± 0.8	~19	Mouse phantom (midcoronal plane)
Audrey Hepburn	1	3 (2, 4, 6 Gy)	551	2.4 ± 2.1	~19	Below 2 mm solid water

Approximate delivery times were based on a 225 kV, 20 mA beam with an output of 0.461 cGy/mAs. The average leaf motion time was ~ 400 ms/aperture, accounting for ~ 20% of total delivery time.

For the first experiment, the SOC’s ability to deliver a 3D concave dose distribution was demonstrated. This type of plan is a hallmark of IMRT and an AAPM standard for IMRT commissioning.⁴⁷ A C-shaped target plan was created for a simple 3D printed block phantom (3.5 × 3.5 × 2 cm, Fig. 9, right) using eight equally spaced coplanar beams, and film was inserted in the center (midaxial plane) of the phantom to analyze the dose distribution perpendicular to the rotational axis.

For the second experiment, A 3D printed mouse phantom (Fig. 9, left) modeled from mouse CT data was used to demonstrate a realistic application for the SOC, as described in Part I. Mouse liver and kidney structures were transferred from a contrast-enhanced mouse CT, and the liver was targeted with a dose of 10 Gy using five beam angles optimized with the 4π algorithm.^48-58 Film was inserted between the two phantom halves for dose measurement.

To demonstrate the ability to optimize and deliver complex 2D dose distributions with the SOC, a third test plan with 1 mm resolution and three dose prescription levels (2, 4, and 6 Gy) was delivered. This plan, optimized to resemble an image of Audrey Hepburn as shown in Part I, includes apertures with equivalent square field sizes ranging from 1 to 25 mm, with a total of 551 apertures and an average aperture size of 2.35 mm. While this plan is arguably more complex than any foreseeable real small animal applications for the SOC, it was developed to visualize the full IMRT capabilities of the system.

For the C-shaped target and mouse phantom liver plans, the maximum, mean, and minimum structure doses were compared between the measured film dose and the calculated, or intended, dose distribution. A pixelwise comparison was performed for the C-shaped plan and mouse liver plan structures to assess the deviation from the intended dose for each pixel, with the maximum and mean absolute deviations reported. Gamma analysis was also performed to compare the calculated and measured dose distributions, for which the typical clinical criteria are 3% dose difference and 3 mm distance to agreement (DTA).⁵⁹ In previous studies, these criteria have been adapted for small animal radiotherapy to 4% and 0.3 mm to account for the approximate order of magnitude difference in imaging resolution.^60,61 Thus, 4%/0.3 mm criteria were used for analysis in this study.

3. RESULTS

3.A. SOC commissioning

The agreement between the SOC and primary collimator axes at the treatment isocenter was within 0.66 mm in the x-direction, and there was no measurable offset in the y-direction. At 1 cm below isocenter, the measured offsets were 0.40 and 0.66 mm in the x- and y-directions, respectively. This suggests an angular misalignment between the SOC and the primary collimator of approximately 1.5° in x and 3.8° in y. These measurements are in close agreement with the Winston-Lutz tests, which showed a difference between the two collimators of 0.53 and 0 mm in x and y, respectively, at the treatment isocenter. Over the full gantry rotation, as shown in Fig. 10 (left), the maximum deviation of the SOC field center from the treatment isocenter was 1.04 mm in x (−1.04 to +0.88) and 1.65 mm in y (−1.65 to +0.1). However, these shifts are reduced when the misalignment of the primary collimator alone is factored in, suggesting that the rotational stability of the SOC itself is within 1 mm in both directions. This rotational stability is also evident in the spoke shot pattern in Fig. 10 (right).

Fig. 10.

(Left) The measured offsets between the sparse orthogonal collimator (SOC) field center and treatment isocenter at each projection angle in the x (blue) and y (red) directions are shown with solid lines. The difference between the offsets with the SOC (mounted on the primary collimator) and the primary collimator alone is shown with dotted lines. (Right) Spoke shot pattern from 1 mm slit SOC fields delivered at eight equally spaced angles, measured perpendicular to the rotational axis and shown with a 2 Gy minimum dose cutoff.

The upper and lower leaf abutting fields patterns are shown in Fig. 11 (left, center). Averaged over six line profiles, the maximum deviations from the average dose were 18.8 ± 3.1% (mean 4.4 ± 0.5%) and 15.5 ± 2.9% (mean 1.9 ± 0.2%) for the upper and lower leaf patterns, respectively. These measurements are within the AAPM Task Group 50⁴⁴ guideline that dose deviations >20% are indicative of leaf misalignment.

Fig. 11.

Abutting field film patterns for the upper (left) and lower (center) leaves, normalized to the average field dose. (Right) One of the alignment verification grid patterns delivered with the sparse orthogonal collimator (SOC), normalized to the maximum dose.

One of the measured grid dose patterns is also shown in Fig. 11. The pixelwise comparison of five sequentially delivered grid patterns showed a 93% match between the 50% isodose distributions, with a maximum deviation of approximately 0.5 mm. Grids compared between three different SOC leaf calibrations showed a pixel match of 91% and a maximum deviation of 0.8 mm.

The midleaf transmission through a single layer of SOC leaves was extremely low, with a maximum measurement of only 0.15 ± 0.05% of the full dose and a mean of 3 × 10–5% (Table II). The interleaf leakage was 0.27 ± 0.05%, which is well below the AAPM recommended maximum of 2%.⁴⁴ There was no statistically significant difference in interleaf leakage between measurements at the four major gantry angles, with a maximum difference of +0.07%, suggesting minimal shifting of the SOC leaf pairs with gantry rotation. The maximum closed leaf transmission, measured along the central matchline of all four leaves, was 7.27 ± 3.40%. The rotational difference in closed leaf transmission was again not statistically significant, with a maximum difference of +1.16%. This leakage is likely due to geometrical imperfections in the machined leaf faces and the 3D printed SOC housing, but is still well within the AAPM recommendation of <25%.⁴⁴ Also, this transmission is irrelevant for SOC plan delivery, since it can be completely mitigated by using off-axis positions for the upper and lower closed leaf positions.

Table II.

Sparse orthogonal collimator (SOC) leaf transmission and penumbra measurements.

	Midleaf (n = 8)	Interleaf (n = 12)	Closed leaf (n = 4)
Transmission (% full field dose)
Maximum transmission (0°)	0.15 ± 0.05% (mean 3 × 10−5%)	0.27 ± 0.22%	7.27 ± 3.40%
Maximum rotational variation (0°–270°)	–	+0.07%	+1.16%

	Upper (n = 12)		Lower (n = 12)
	Tongue	Groove	Tongue	Groove	Average
Penumbra (80 to 20% dose distance, mm)
Leaf edge	0.79 ± 0.04	0.81 ± 0.06	0.82 ± 0.04	0.75 ± 0.07	0.79 ± 0.05
Leaf face	0.80 ± 0.05		0.80 ± 0.1		0.80 ± 0.08
Square field size (n = 4):	26 mm	20 mm	10 mm	5 mm	Range
2 mm depth	0.86 ± 0.09	0.97 ± 0.10	0.80 ± 0.07	0.80 ± 0.13	0.17
20 mm depth	1.56 ± 0.24	1.26 ± 0.25	1.15 ± 0.14*	1.00 ± 0.12**	0.55

^*Significantly different from the 26 mm field penumbra (two-tailed t test, 5% significance level).

^**Significantly different from the 10 and 26 mm field penumbras (two-tailed t test, 5% significance level).

The leaf penumbra measurements are also shown in Table II, and the field profiles for a range of square field sizes are shown in Fig. 12 for reference. There was no significant difference in penumbra between upper and lower leaf faces, upper and lower leaf edges, or tongue and groove leaf edges. The average leaf edge and face penumbras were 0.79 ± 0.05 mm and 0.80 ± 0.08 mm, respectively, which also had no significant difference. This is beneficial because although leaf edges are not used to form aperture boundaries in a typical SOC plan, they could theoretically be used without resulting in an asymmetrical field penumbra (e.g., the delivery of two diagonal quadrants with a single leaf layer). For measurements at 2 mm depth, there was no significant difference in penumbra with leaf position for square apertures 5 to 26 mm, with a range of 0.17 mm. There was slightly more variation in penumbra at 20 mm depth, with a range of 0.55 mm and significant differences between several of the field sizes (5 vs 10 mm, 5 vs 26 mm, and 10 vs 26 mm). However, it should be noted that the source-to-leaf distance of the SOC was extended after machining of the current leaves in order to achieve superior dosimetry at the treatment isocenter. Therefore, the radius of curvature of these leaves is slightly less than it would be for a fully double-focused system, which would theoretically result in a more consistent leaf penumbra. However, based on the system and setup geometry, the measured penumbra range at 2 mm depth suggests a focal spot size of 2.6 to 3.1 mm. Since the reported diameter is approximately 3 mm, the large focal spot is likely the main source of the measured penumbra.

Fig. 12.

Normalized lateral field profiles for square field sizes of 1 to 25 mm, measured at a depth of 2 mm in solid water.

3.B. Dosimetric testing

Figure 13 shows the calculated (left) and measured (center) dose distributions for the C-shaped target plan. The gamma analysis with 4%/0.3 mm criteria revealed a pass rate of ~95% for pixels within the target structure and 85% for the entire field shown. The maximum and mean absolute pixelwise dose differences were 4.12 and 0.59 Gy, respectively (Table III), and the measured dose to the target had slightly higher maximum (15.8% of the prescription dose), mean (7.0%), and minimum (13.5%) doses. The 50% isodose lines are shown in Fig. 13 (right), demonstrating excellent overall agreement between the calculated and measured dose distributions.

Fig. 13.

(Left) Calculated dose distribution of the C-shaped target plan perpendicular to the gantry rotation axis. (Center) Measured film dose distribution from the center of the solid water phantom for the C target plan delivered with the sparse orthogonal collimator. Both plans are shown with the same color scale, in units of Gy. (Right) A comparison of the calculated (yellow) and measured (blue) 50% isodose lines, with overlapping regions shown in red.

Table III.

Comparisons between the measured and intended dose distributions for the C-shaped target plan and the mouse phantom whole liver plan.

		Dose statistics (measured – calculated)			Pixelwise dose comparison
Plan	Structure	Max	Mean	Min	Max diff	Mean diff	Gamma pass rate
C target	C	+1.58 (15.8%)	+0.70 (7.0%)	+1.35 (13.5%)	4.12 (41.2%)	0.59 (5.91%)	94.9%
Mouse LIVER	Liver	+1.30 (13.0%)	−1.02 (10.2%)	+1.10 (11.0%)	3.50 (35.0%)	1.19 (11.9%)	98.2%
	Kidneys	+0.43 (4.3%)	+0.24 (2.4%)	+0.11 (1.1%)	0.43 (4.3%)	0.24 (2.4%)	100%

Max, mean, and min dose differences written as [Gy (% prescription dose)]; max and mean diff are absolute pixelwise dose differences; gamma analysis was performed with 4%/0.3 mm criteria for dose/distance.

The results of the mouse phantom liver test plan are shown in Fig. 14, with the liver and kidney dose comparisons given in Table III. The maximum measured liver dose was 13.0% higher than the calculated dose, the mean was 10.2% lower, and the minimum was 11.0% higher. As evident in the film dose distribution and isodose comparison shown in Fig. 14(c) and (d), the lower left portion of the liver was cutoff due to slight phantom misalignment. The affected pixels were omitted from the liver dose analysis, and the measured dose distribution was rotated slightly to account for the setup error. For the unaffected pixels within the liver, the gamma analysis showed a high pass rate of 98.2%. The measured SOC plan was able to significantly spare the dose to the kidneys, with maximum and mean doses of 0.43 and 0.24 Gy, respectively. These are only 4.3% and 2.4% higher than the calculated doses, and therefore, all pixelwise differences for the kidneys were within 5% of the intended dose.

Fig. 14.

(a) Midcoronal view of the calculated dose for the mouse phantom whole liver plan (units of Gy). (b) The 5 optimal coplanar beam angles selected with the 4π algorithm. (c) Measured film dose from the mouse phantom, treated with the whole liver plan, at the plane shown in A (units of Gy). (d) A comparison of the calculated (yellow) and measured (blue) 60% isodose lines, with overlapping regions shown in red. Target structure was rotated to account for slight phantom misalignment, which also resulted in the truncated lower left portion of the target.

The calculated and measured doses for the 2D Audrey test plan are shown in Fig. 15. The maximum and minimum measured film doses were both 1.1 Gy higher than the calculated dose distribution (12.2% of the maximum intended dose), with a mean pixelwise absolute dose difference of 1.6 Gy. Although this plan shows some discrepancies in absolute dose prediction for very small apertures sizes, the spatial distribution is extremely similar to the calculated plan, validating the overall accuracy of the SOC hardware and control software. The sources of this absolute dosimetric error are as discussed in Part I due to the uncertainties in determining the small field output and off-axis factors, which are extremely sensitive to SOC alignment relative to the x ray source. Currently, SOC is removed for imaging, which is needed to setup the phantoms and then installed for IMRT, resulting in varying small field factors that have a noticeable cumulative error in IMRT delivery. We expect substantially reduced error with a permanent SOC mounting solution by the vendor.

Fig. 15.

(Left) Calculated Audrey test plan with 4 dose levels and an average aperture size of 2.35 mm. (Right) Measured dose distribution of the Audrey plan delivered with the sparse orthogonal collimator. Both plans are shown with the same color scale, in units of Gy.

4. DISCUSSION

While Part I of this report demonstrates the planning capabilities of the SOC system for preclinical x ray IMRT, Part II demonstrates the engineering feasibility, fabrication, and delivery performance. Compared to the MLC, we believe that the SOC is better suited for preclinical applications. First of all, SOC resolution is independent of the leaf width in both collimation directions, avoiding the engineering challenge of further reducing the MLC leaf width. Consequently, the SOC has substantially fewer moving components than the MLC, making miniaturization feasible and more robust. SOC modulation resolution depends on leaf step size, which is determined by the stepper motors, and can therefore in theory be on the order of a micron. In practice, this mechanical resolution is limited by the resolution of the motor encoders and the accuracy of the housing fabrication, but is still well below a millimeter for the described SOC design. The current RAO formulation described in Part I, however, does have a finite beamlet resolution due to the discretized nature of the rectangles, which does not take full advantage of the resolution capabilities of the SOC geometry. Further investigation of a multiresolution approach must be conducted to better exploit this design.

The SOC design is also advantageous over a conventional MLC because it makes closed leaf-end transmission inconsequential. While this is significantly lessened with a dual-layer MLC, this configuration still has a region of higher transmission where the closed leaf matchlines intersect. These matchlines are not necessary with the SOC, since it is possible to block the entire SOC field without even fully closing opposing leaf pairs. The maximum transmission in this SOC configuration would be from the photons in the center point of the field, passing through both upper and lower tongue and groove joints, which would only be approximately 0.07% based on our interleaf transmission measurements. In general, these low transmission levels are an advantage of the SOC leaf design, with our measurements suggesting an average midleaf transmission through both layers of only about 0.02%.

Although the inherent delivery efficiency of the SOC design is lower than that of a typical MLC system, the ability to deliver two fields in diagonal quadrants simultaneously can potentially reduce delivery times by up to 50% compared to jaws-only IMRT. These reductions are even greater when we consider the added time from off-axis aperture couch shifts with a system like the motorized variable collimator. Also, the SOC aperture optimization can be easily tuned to select fewer, larger apertures for plans where short delivery times are a priority. Additionally, although currently the leaf repositioning time is ~400 ms/aperture and accounts for approximately 20% of the overall delivery time, we believe there is significant room for reduction with more optimized leaf motion control.

Overall, the preliminary results from the SOC are promising. The SOC leaves have a very consistent penumbra, varying by only 0.17 mm with leaf position at 2 mm depth and with no measurable distance between the upper and lower sets, despite the current discrepancy between the beam divergence and leaf radii. Using just the quick, simple leaf position calibration procedure described, abutting field patterns were delivered with <20% dose difference at the matchlines, as recommended by AAPM guidelines.⁴⁴ In the C-shaped target experiment, the SOC achieved a highly concave dose distribution that agreed well with the intended 50% isodose area and had a gamma pass rate of 94.9% within the target for 4%/0.3 mm criteria. An end-to-end test of a realistic SOC application was performed with the mouse phantom whole liver plan, demonstrating all steps from the mouse CT scan to the final treatment delivery. With this plan, a conformal dose was delivered to the mouse liver while significantly sparing dose to the kidneys. The measured and intended target doses were again in close agreement, with a gamma pass rate of 98.2%. The ability to create and deliver a highly complex dose 2D distribution was also demonstrated with the Audrey plan.

However, there is still a significant degree of error and uncertainty with the current SOC system. In addition to the dosimetric error discussed in Part I, there are also various sources of geometrical uncertainty. The first is misalignment of the SOC with the primary collimator, which we measured to be approximately 0.7 mm at a gantry angle of 0°. Combined with the isocenter walk due to the flex of the SOC and slight play between the SOC and the primary collimator, this resulted in measured deviations from the treatment isocenter of up to 1.65 mm for a full gantry rotation. There is also uncertainty in the leaf calibration procedure, with leaf position deviations of up to 0.8 mm observed between calibrations. Leaf position repeatability measurements with the same calibration also showed differences of up to 0.5 mm. Since this is well below the resolution of the encoders, it is more likely due to slight shifting of the motors within the SOC housing or inconsistencies in the leaf movement relative to the motors.

The largest source of geometrical error is currently the subject setup. Since the treatment planning process is not yet fully automated and planning cannot be realistically performed with the subject on the couch, the subject setup in the planning CT must be recreated for treatment delivery. Although this is standard practice in clinical treatment, the SmART system software lacks the highly automated tools used for reproducible patient setup in the clinic. Subject setup for SOC treatment is a much more manual process with greater uncertainty, likely on the order of a couple millimeters or degrees of misalignment. Even with a fast, fully integrated treatment planning system, this error will not be completely mitigated until a more permanent SOC installation is possible, since the SOC must currently be reinstalled and recalibrated between imaging and treatment. It should also be noted that all measurements (apart from the Winston-Lutz test) were performed with EBT3 film, which has an inherent uncertainty of ~2%^62,63 in dose and was limited by our scanning resolution of 96 dpi, which was used by default but could be increased for higher resolution analysis, at the cost of increased noise.

Although these sources of geometrical and dosimetric error add up to significant uncertainties with current SOC treatment, the majority of this error could be substantially reduced with integration into the SmART system. The x ray beam is currently controlled with the console during SOC treatment, requiring the user to manually start and stop the beam. Integration with the SmART software would enable automated beam control, which is especially beneficial for pausing the beam if any hardware errors are detected. This would also allow us to develop a more accurate, automated calibration procedure using the projected leaf positions on the flat panel detector. Permanent installation on the SOC gantry, or at least a custom adapter to bypass the primary 4 × 4 cm collimator attachment, could significantly reduce the treatment isocenter misalignment and the potential for setup error between imaging and treatment.

5. CONCLUSIONS

The sparse orthogonal collimator was designed as a simple, more practical dose-modulation device for small animal IMRT. This SOC design is driven by a novel direct aperture optimization algorithm that uses only rectangular apertures for complex dose modulation. With the complete SOC hardware and commissioned planning system, we demonstrated extremely low leaf transmission, consistent penumbra, and the ability to deliver conformal, complex dose distributions in close agreement with intended treatment plans. This novel system could considerably enhance our ability to perform clinically translatable animal radiation research in both cancer- and non-cancer-related fields.