Can parametric software standardize collimator creation for preclinical irradiation studies?

What the parametric workflow actually generates

Collimator Creator (CC) is an open-source parametric workflow that converts aperture dimensions and jig offsets into printable STL models. Each design produces four functional solids: a collimator mold, a collimator shell, a Cerrobend holder, and a beam degrader. The meshes are watertight, meaning a closed surface suitable for slicing or geometry import, so they can go straight to fused deposition modeling or into Monte Carlo geometry.

The physical stack follows the beam path. A 10 mm PLA beam degrader sits upstream of the aperture and sets the mean electron energy and scatter incident on the block. The Cerrobend collimator, 15 mm along the beam in this work, is cast from a disposable PETG mold; the mold is only a casting fixture, while the alloy defines projected field size and penumbra at the specimen. A thin PLA collimator shell registers the degrader and the Cerrobend collimator to the custom electron applicator, and a separate PLA holder supports the demolded alloy.

The interface maps numerical inputs directly to dosimetric degrees of freedom: width and length for rectangular apertures, a single radius for circular apertures, or a custom outline including contour-derived shapes with margin. The aperture center aligns with the central axis by default, and users can enter crossline and inline offsets in millimeters. Only the two rectangular lung fields were dosimetrically evaluated, so the circular and contour capabilities are software features rather than validated geometries.

Two lung fields defined by one offset parameter

The study fabricated and characterized two designs. The bilateral lung collimator is 4.0 x 1.5 cm2 and centered on the central axis. The unilateral lung collimator is 1.5 x 1.5 cm2 with a 7.5 mm jig offset. Both use a 15 mm thick Cerrobend collimator and a 10 mm PLA degrader.

The point of the offset is that the collimator shell displaces the specimen jig relative to the central axis, so a single block geometry can serve bilateral (centered) versus unilateral lung irradiation with a controlled lateral shift. That is the concrete meaning of "parametric" here: the aperture shape and the specimen position are both numerical inputs, not hand-cut adjustments.

Fabrication followed a defined chain. STL meshes were sliced in BambuSlicer 1.9.7 and printed on a Bambu Lab X1 Carbon FDM printer. The mold was printed in PETG because its glass-transition range of 79-85 degrees C exceeds the 70-75 degrees C melting range of Cerrobend. The holder, shell, and degrader, which are not exposed to molten alloy, were printed in standard PLA. Cerrobend was melted to 75 degrees C and poured in a slow, continuous stream to minimize air entrapment, cooled in still air for at least 30 minutes, then demolded with taps on the aperture wall with a rubber mallet.

Casting accuracy and uniformity checks

Kilovoltage planar images were used to assess aperture geometry and casting uniformity. Images were acquired with the on-board kV imaging system of the TrueBeam at 50 kVp and 1.6 mAs (40 mA, 40 ms), and length and width were measured on the image projected to the isocenter plane using the source-to-axis to source-to-image distance ratio.

Comparisons between CAD dimensions and the fabricated Cerrobend collimators showed deviations of less than 0.2 mm for both geometries. For the unilateral collimator, the central 90% region of interest, after excluding the aperture and a 1-mm margin around it, yielded a uniformity of 99.1%, with no pixels deviating more than 5% from the mean open-aperture signal. The 1.5 x 1.5 cm2 collimator was chosen for this defect check because it contained the largest Cerrobend-covered area, maximizing the area evaluated for potential defects.

These are geometric and casting-quality measures, not dose measures. They establish that the printed mold and the poured alloy reproduced the intended aperture, and that no observable Cerrobend defects appeared in the sampled region. They do not by themselves establish how dose is distributed at the specimen.

FWHM and penumbra under two dose rates

Radiochromic film was positioned flush with the downstream face of the collimator to record the in-air fluence map, stabilized for 24 hours, and scanned at 150 dpi (0.169 mm pixel spacing) with color corrections disabled. The red-channel optical density was converted to dose using a pre-established calibration curve, and one-dimensional profiles along the principal X and Y axes were extracted. FWHM was calculated, and the 80%-20% penumbra was computed at each field edge with the left and right values averaged.

For the bilateral lung collimator (4.0 x 1.5 cm2) at conventional dose rate, measured FWHM was 4.10 cm in x and 1.54 cm in y; under UHDR the values were 4.09 cm and 1.53 cm. Mean 80%-20% penumbra at conventional dose rate was 0.89 mm in x and 1.03 mm in y; under UHDR it was 0.73 mm and 0.90 mm. For the unilateral lung collimator (1.5 x 1.5 cm2) at conventional dose rate, FWHM was 1.52 cm in x and 1.53 cm in y; under UHDR, 1.52 cm and 1.53 cm. Penumbra at conventional dose rate was 1.03 mm in x and 1.05 mm in y; under UHDR, 0.90 mm and 0.92 mm.

All FWHM values are within 3% of their designed widths, and penumbra across both collimators and both delivery modes ranged from 0.73 to 1.05 mm. The paper is explicit that these measurements compare field widths under conventional and UHDR delivery but do not test a dose-rate effect. FWHM and penumbra answer different questions: FWHM describes how wide the field is at half of central-axis dose, while penumbra describes how sharply dose falls off at the edge. A field can match its nominal width and still have a wider or narrower edge falloff.

Off-axis leakage at conventional dose rate

Off-axis leakage was measured with an ionization chamber (PTW 30010) at conventional dose rate only. The setup placed 1 cm of solid water upstream and 1 cm downstream, both simulating mouse tissue, with a 2 cm air gap between phantom and collimator, and each position was measured three times. Electrometer background charge was subtracted, and a readout threshold excluded residual background signal. Off-axis percentage leakage doses were calculated relative to the central-axis dose, with uncertainty taken as the standard deviation of the three paired ratios.

For the bilateral lung collimator, relative leakage along the crossline at 4 cm from the central axis, which is 2 cm from the collimator edge, was 1.6226 +/- 0.0002%. For the unilateral lung collimator, relative leakage at 3.5 cm from the central axis, also 2 cm from the collimator edge, was 1.3102 +/- 0.0002%. Both are below 2%.

Because leakage is expressed as a ratio of two readings acquired with the same chamber, electrometer, phantom, and beam configuration, the chamber calibration coefficient and its influence quantities cancel and do not contribute to the reported uncertainty. That is why the stated uncertainties are so small: they reflect repeatability of the ratio, not the full calibration chain. Leakage was not measured under UHDR delivery.

What the validation does not cover

The validation is limited to two rectangular fields measured with film at the downstream face of the collimator. It does not establish output factors, surface-dose perturbation, or dose at the specimen location. Circular and custom apertures were not dosimetrically validated, and fabrication time, inter-user variability, and manufacturing reproducibility were not compared with a manual workflow.

The paper also does not test a dose-rate effect, and it does not assess the biological FLASH effect. Platform-level beam characterization and UHDR dosimetry were reported previously and were not re-evaluated here. Monte Carlo integration is described as a future direction rather than a validated component: CC exports standard STL files that could be imported into Monte Carlo radiation transport simulations, but no simulation, simulation-to-measurement comparison, or inverse design was performed.

The tool currently produces static apertures of uniform thickness. Variable-thickness collimators or 3D-printed compensators would require separate dosimetric validation. Automatic generation from a delineated target was also not tested; users specify dimensions or upload contours manually.

Why digital design inputs matter more than print speed

The paper frames its contribution against hand-crafted Cerrobend blocks, where casting and cutting are prone to human error. It cites a comparison in which manually fabricated Cerrobend inserts deviated from planned field outlines by 1-3 mm, whereas 3D-printed alternatives achieved alignment within 0.4 mm. The present study did not directly compare fabrication time, inter-user variability, or manufacturing reproducibility with a manual workflow, so its own reproducibility claim rests on a different property: CC defines collimator geometries programmatically, so identical STL models can be produced and shared from the same inputs.

That distinction matters for how the result should be read. The measured sub-0.2 mm deviation shows that this pipeline reproduced its own specified designs for two geometries. It does not show that the pipeline is faster than manual fabrication, or that different users would produce identical results, because those comparisons were not run. The reproducibility argument is about the design file being deterministic, not about the full fabrication and dosimetric process being reproducible end to end.

The paper also notes that many UHDR-capable linac beamlines lack dedicated field-shaping accessories, so investigators rely on bespoke Cerrobend blocks. A digital STL can be distributed electronically so researchers at different sites can print the same accessory, which is the mechanism by which a validated design library could support multi-institutional use of the same beam geometries. Design transfer across institutions was not evaluated in this study.

The priority is geometry, not biology

The most decision-relevant result in this paper is the pairing of two numbers: cast apertures within 0.2 mm of design, and FWHM within 3% of nominal field dimensions for both the 4.0 x 1.5 cm2 and 1.5 x 1.5 cm2 fields under both conventional and UHDR delivery. Together they say that for these two rectangular electron fields, the parametric route delivered the intended geometry at the downstream collimator face.

The boundary is equally specific. Penumbra ranged from 0.73 to 1.05 mm, and leakage 2 cm beyond the collimator edge stayed below 2% at conventional dose rate only. Nothing here establishes dose at the specimen, output factors, or surface-dose perturbation, and no dose-rate or FLASH effect was tested. A reader deciding whether to adopt this workflow should treat it as validated for rectangular linac-based electron fields of the tested sizes and offsets, with the specimen-level dosimetry still to be done.

The paper's own next steps point the same way: specimen-level dosimetry, additional aperture geometries, and Monte Carlo integration. Until those exist, the value of CC is that it standardizes the digital design input, not that it has been shown to standardize the delivered dose.

Sources used in this answer