A multicentre audit of the geometric accuracy and water equivalence of 3D printed thermoplastic objects for use in radiotherapy☆ Authors: Matthew R. Jones, Peter D. Woolliams, Matthew A. Bolt, Owen McLaughlin, Maria Boutros, Tristan Wright, Rhys Jenkins, Anna Tonino, Fiona Milliken, Neil Bentley, Gordon D. Sands, James C.L. Burnley, Paul J. Doolan, Dualta McQuaid, Geoff J. Budgell, Elizabeth J. Adams, Conor K. McGarry, Catharine H. Clark Published: 2026 ABSTRACT Graphical abstract Introduction Increasing adoption of 3D printing techniques in radiotherapy has seen fabrication of dosimetric and patient-specific phantoms, brachytherapy applicators, bolus structures, patient immobilization and proton beam compensators [1], [2], [3], [4], [5]. Water is the reference material for radiotherapy dosimetry, and plastics are commonly substituted when water use is impractical [6], [7], [8]. There is therefore a need to validate the water equivalence of thermoplastic polymers used in 3D printing. Fused deposition modelling (FDM) printers vary object density by defining internal proportions of thermoplastic and air. This 'infill' percentage significantly affects dosimetric properties and studies have characterised thermoplastics, demonstrating linear relationships between infill and CT number [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19]. Polylactic acid (PLA) was selected for this work due to its tissue-equivalent properties and widespread clinical use [2], [20]. Alternative thermoplastics, acrylonitrile–butadiene–styrene (ABS) and glycol-modified polyethylene terephthalate (PETG), have also demonstrated tissue-equivalent properties, but are less commonly used and associated with reduced printing stability [21], [22], [23], [24]. This paper presents a multicentre audit designed to quantify the geometric accuracy obtainable with FDM printers and the variations in physical and dosimetric properties of printed PLA objects, to support consistency and guide practical implementation in radiotherapy. Materials & methods Audit format Three cuboid blocks were designed using the computer-aided design (CAD) package Fusion 360 v2.0.17710 (Autodesk Inc. California, U.S.A.), with dimensions 7.0 cm × 7.0 cm × 0.8 cm (block A), 7.4 cm × 7.4 cm × 1.0 cm (block B) and 7.8 cm × 7.8 cm × 1.2 cm (block C). Dimensions allowed for a range of measurements and accommodated a 5 × 5 cm photon beam. Block A also featured a range of protruding and recessed discs (referred to as 'hills' and 'holes' respectively), positioned on the top (horizontal) and side (vertical) faces and ranging in diameter from 1 to 6 mm in 1 mm steps. A 12 cm × 12 cm × 3 cm attenuation phantom was printed, with a channel shaped to accommodate a PTW Semiflex ionisation chamber (PTW Freiburg GmbH) with its physical centre at a depth of 2 cm from the top face. A jig was also designed to hold the blocks vertically for CT scanning purposes (Fig. 1). Blocks A–C, the CT jig and the chamber phantom are provided as stereolithography (STL) files via an external dataset repository [25].Fig. 1Test objects on attenuation phantom (left) and in CT jig (right). The nomenclature used in this paper defines the ‘top’ of block A as the face bounded by the edges labelled ‘length’ and ‘width’, and the ‘side’ as that bounded by ‘length’ and ‘height’. Test objects on attenuation phantom (left) and in CT jig (right). The nomenclature used in this paper defines the ‘top’ of block A as the face bounded by the edges labelled ‘length’ and ‘width’, and the ‘side’ as that bounded by ‘length’ and ‘height’. The attenuation phantom was printed at the coordinating centre at 100% infill. The three blocks and CT jig were provided to audit participants as STL files. Blocks were converted to G-code format using the participants’ chosen software and printed at 100, 90 and 80% infill for blocks A, B and C respectively. The use of a 'cubic' infill pattern, which fills the object with cubes positioned vertically on one corner, or similar (depending on software) was recommended for consistency. Printers were otherwise operated using routine local settings, calibration and quality assurance (QA), to reflect current practice. Blocks and data from ten centres were audited, with details of equipment, PLA type, and print settings recorded (Table S1, Supplementary Material A). The printed blocks were CT scanned with their largest faces oriented vertically in the axial plane, using a tube potential of 120 kV; other settings followed the participants’ standard head and neck protocols. This orientation minimised interpolation effects [26] and reduced statistical noise by accommodating larger regions of interest (ROIs). Use of the CT jig was optional, with some centres supporting the blocks using radiolucent patient supports. Analysis was performed by a single operator at the coordinating centre. All data were reported as mean ± standard deviation (SD) unless otherwise specified. Normality was assessed using Shapiro–Wilk. Parametric (ANOVA) or non-parametric (Mann–Whitney U) tests were applied as appropriate, with significance set at α = 0.05. Geometric accuracy The geometric accuracy of the blocks was measured using a calibrated RS PRO digital calliper (accuracy 0.03 mm) at several points along each edge and around the diameter of each disc. The discs were measured at their lowest point (the base of the hills and the bottom of the holes) for stability and the maximum and minimum measurements recorded for each. As the 1 mm holes were too small to accommodate the callipers, diameters were measured at the surface by determining the distance between opposing edges of each hole. CT numbers All blocks were scanned at the coordinating centre using the standard head and neck imaging protocols on the reference CT scanner, a GE Discovery (GE Medical Systems, Waukesha, WI). Blocks were positioned vertically in the CT jig, centred on the middle of block B and scanned with a slice thickness of 1.25 mm (50 cm field of view, 120 kV, automated mA, helical acquisition). Images were imported into the Eclipse treatment planning system (TPS), (Varian Medical Systems, Palo Alto, USA) along with those taken at the participant centres. A 5 × 5 cm square ROI was centred within each block on the axial slice closest to the geometric mid-plane using on-screen measurement tools, and the mean and SD of the Hounsfield units (HU) were recorded. Nominal infills were plotted against the measured HU densities and the infill corresponding to water equivalence (%WE) for each centre was calculated by interpolating linear trendlines to zero HU. This was completed for both participant and reference CT images for comparison. Mass density The masses of all blocks were measured using a precision balance, calibrated to an accuracy of 0.03 g at the measurement range (Kern & Sohn GmbH, Balingen, Germany). These were used alongside the average of the maximum and minimum measured dimensions for each block to calculate the mass densities. To remove the potentially confounding influence of the outer walls, the measured dimensions of blocks B and C were adjusted for the top and bottom layer and wall thickness settings provided by the participants. For block A (100% infill), the hills and holes had matched nominal dimensions so their net volumes were assumed to cancel and the blocks were treated as regular cuboids when calculating volume. The masses of the combined walls were estimated using the calculated mass density of the 100% infill block and subtracted from the measured masses of blocks B and C. This was then used to estimate the densities of the inner volumes of the blocks containing the assigned infills. Mass densities were then plotted against the measured CT numbers and the mass densities corresponding to water equivalence (ρWE) were calculated by interpolating linear trendlines to zero HU. Attenuation For each configuration of blocks, the HU values were converted to relative electron densities (RED) using the calibration curve for the reference CT. The combined water equivalent thickness (WET) was then estimated as the radiological depth defined in AAPM report 85 [27] (Equation (1)):(1)WET=∑i(REDi×di)where REDi and di are the RED and physical thickness of the ith layer of PLA in the beam path. The top and bottom layers of the blocks were assumed to be 100% infill with thicknesses rounded up to integer multiples of the specified layer heights. A PTW Semiflex ionisation chamber was inserted into the cavity of the attenuation phantom and positioned at the isocentre of a Varian TrueBeam® linear accelerator (Varian Inc. Palo Alto, U.S.A.). The blocks were positioned on top of the phantom in configurations placing the chamber at nominal depths of 20 mm to 50 mm and irradiated using 5 × 5 cm 6 MV photon fields. A parameter TPReff (an effective tissue phantom ratio normalised to the nominal 20 mm depth in the attenuation phantom) was calculated as the ratio of the average chamber measurement taken through each block configuration to that with no blocks present. These were then plotted against the calculated WETs for all centres. Golden Beam Data (GBD) for TrueBeam 6 MV photons (Varian Inc. Palo Alto, U.S.A.) were used to generate reference data for comparison. Depth-dose data were converted to TPRs for a field size of 5 × 5 cm and reference depth of 20 mm, using the process described in BJR Supplement #25 Appendix B [28] (Supplementary Material B). Results All blocks printed smaller than intended, with a mean dimensional error of (−0.2 ± 0.3) mm (Fig. 2). All measured errors were normally distributed between centres (p = 0.17–0.74). The apparently larger width errors were not significantly different to the height and length errors, although this result was marginal (p = 0.05). The mean deviation from the expected diameters of all discs was also (−0.2 ± 0.3) mm, with errors predominantly negative (Fig. 3). Data grouped by nominal diameters were normally distributed (p = 0.06–0.17). Mean diameters did not differ significantly between the 2–6 mm features (p = 0.73), but inclusion of the 1 mm features produced a significant difference (p < 0.001) due to the 1 mm holes printing smaller than intended at both orientations (mean diameter = (0.5 ± 0.3) mm; Fig. 3c,d). Hills on the top faces were significantly smaller than all other disc types and locations (p < 0.001; Fig. 3a). No significant differences were observed between the remaining groups (p = 0.18–0.98; Fig. 3b–d).Fig. 2Measured differences from nominal height, length and width of blocks A, B and C for all centres. Boxes show interquartile range (IQR) with median lines, whiskers extend to the most extreme values within, and outlying points are plotted as dots. Means are denoted by 'x'.Fig. 3Differences in maximum and minimum diameters of disc features, compared to intended diameters, by nominal size for (a) hills on the top face, (b) hills on the side edge, (c) holes on the top face and (d) holes on the side edge. Nominal diameters are indicated on the x-axis. Boxes show IQR with median lines, whiskers extend to the most extreme values within, and outlying points are plotted as dots. Means are denoted by 'x'. Measured differences from nominal height, length and width of blocks A, B and C for all centres. Boxes show interquartile range (IQR) with median lines, whiskers extend to the most extreme values within, and outlying points are plotted as dots. Means are denoted by 'x'. Differences in maximum and minimum diameters of disc features, compared to intended diameters, by nominal size for (a) hills on the top face, (b) hills on the side edge, (c) holes on the top face and (d) holes on the side edge. Nominal diameters are indicated on the x-axis. Boxes show IQR with median lines, whiskers extend to the most extreme values within, and outlying points are plotted as dots. Means are denoted by 'x'. Measured CT numbers varied between participants by up to 120 HU for 80% infill, 200 HU for 90% infill and 180 HU for 100% infill. HU data grouped by infill were normally distributed (p = 0.062), with all data points within ± 2 SD of the mean. Calculated inner densities varied by up to 0.15, 0.20 and 0.14 g/cm3 for 80%, 90% and 100% infill respectively. The blocks with 100% infill had densities of 1.11–1.26 g/cm3. Inner density and CT number showed a strong positive correlation (r = 0.981, p < 0.001; Fig. 4). There were no significant differences in the %WE and ρWE values calculated using the reference and participants' CT images (p = 1.00 and 0.84 respectively; Table 1; Fig. S1, Supplementary Material C). Data from centres using PLA formulations containing additional strengthening ingredients (PLA+, Tough PLA) fell both above and below the overall means and remained within ± 2 SD (Table S3, Supplementary Material E). Density and dosimetric data for Centre 7′s block B were excluded because the block was inadvertently printed at 80% infill.Fig. 4Calculated inner densities plotted against measured CT density. Error bars represent 2 SD from the mean HU (vertical) and the propagated uncertainties (k = 2) based on the achievable accuracy of the geometric and mass measurements (±0.03 mm and ± 0.03 g respectively). The data from each centre, indicated by colour/shape of the plotted points, represent printed infills of 80%, 90% and 100% read left to right.Table 1Mean water-equivalent infill and mass densities calculated using linear fits to reference and participant CT image data.CT scanner%WE (%)ρWE (g/cm3)Mean ± SDrangeMean ± SDrangeReference CT94.5 ± 3.589.3–100.11.09 ± 0.031.06–1.15Participant CTs94.4 ± 3.788.7–100.81.10 ± 0.021.06–1.14Both94.5 ± 3.588.7–100.81.10 ± 0.031.06–1.15 Calculated inner densities plotted against measured CT density. Error bars represent 2 SD from the mean HU (vertical) and the propagated uncertainties (k = 2) based on the achievable accuracy of the geometric and mass measurements (±0.03 mm and ± 0.03 g respectively). The data from each centre, indicated by colour/shape of the plotted points, represent printed infills of 80%, 90% and 100% read left to right. Table 1Mean water-equivalent infill and mass densities calculated using linear fits to reference and participant CT image data.CT scanner%WE (%)ρWE (g/cm3)Mean ± SDrangeMean ± SDrangeReference CT94.5 ± 3.589.3–100.11.09 ± 0.031.06–1.15Participant CTs94.4 ± 3.788.7–100.81.10 ± 0.021.06–1.14Both94.5 ± 3.588.7–100.81.10 ± 0.031.06–1.15 Mean water-equivalent infill and mass densities calculated using linear fits to reference and participant CT image data. The PLA data followed the water reference curve closely, with a maximum deviation of 0.6% (Fig. 5a), and all data lay within ± 2 SD of the mean values at each depth (Fig. 5b). Calculated WETs for each nominal PLA thickness varied by 1.2–3.0 mm ((1.9 ± 0.7) mm; Table S2, Supplementary Material D), while measured TPReff values varied by 0.005–0.011 (0.007 ± 0.002) across centres. These differences increased approximately linearly with thickness, with the maximum inter-centre TPR difference reaching 1.2% at the nominal 50 mm depth.Fig. 5(a) TPReff values compared with calculated 6 MV TPR data for water; PLA data follow the water reference curve closely, with a maximum deviation of 0.6%. (b) TPReff data plotted against water‑equivalent depth (WET) in PLA. Points are colour‑coded by nominal WET, and error bars represent ± 2 SD. (a) TPReff values compared with calculated 6 MV TPR data for water; PLA data follow the water reference curve closely, with a maximum deviation of 0.6%. (b) TPReff data plotted against water‑equivalent depth (WET) in PLA. Points are colour‑coded by nominal WET, and error bars represent ± 2 SD. Discussion This multicentre audit assessed geometric and dosimetric properties of 3D-printed PLA objects across ten centres. Results showed achievable geometric accuracy and physical densities were broadly consistent with previous studies, despite inter-centre variation. PLA behaved predictably as a tissue substitute, facilitating local QA and reference measurements, and providing a basis for standardising tolerances, print design, and QA procedures. Pooled data from the literature regarding 3D printing errors using FDM and PLA reported combined mean errors of (−0.2 ± 0.1) mm across studies reporting error direction, and (0.2 ± 0.1) mm absolute error across all studies (Table S4, Supplementary Material, [44], [45], [46], [47], [48] F). These were consistent with audit results, suggesting a general tolerance inherent to FDM-printed PLA phantoms. Recessed 1 mm holes and protruding discs on the top faces were shown to be more prone to geometric errors than details of other type and orientation. These errors could adversely affect the intended use of printed items, for example in cases where 1 mm channels or imaging resolution details are required. In such cases, the data presented in Fig. 3 can inform mitigation at the design stage, such as adding 0.5 mm to hole diameters. While the smaller width measurements indicated by Fig. 2 were not found to be statistically significant, this result was marginal and negative deviations predominating along this axis have been reported elsewhere [[29], [30]]. All printers used for this audit were of rectilinear design, with the horizontal movement of the nozzles driven by tension belts or leadscrews. Geometric accuracy may be influenced by printer axes and the physical characteristics (belt tensions, motor calibration etc.) that affect them. Sources of error have been investigated, with layer thickness, printing speed, thermal shrinkage and the anisotropic nature of the FDM process identified as significantly impacting accuracy [[29], [31], [32], [33]]. Audit data provided information on achievable accuracy that can inform QA tolerances, and optimisation of print design and orientation. Given the scope for inter-centre variation, routine QA testing of 3D-printed objects used in radiotherapy should be considered standard practice. In addition to following manufacturer recommendations for printer maintenance, variations can be reduced through appropriate material storage, control of print environment, and verification of printed dimensions and density in line with recent IPEM guidance [34]. The larger deviations for 1 mm features reflect the resolution limits of FDM with standard nozzles. Where 1 mm detail is required, CAD oversizing and manual post-processing corrections should be considered. Consistency between prints was previously investigated, with Craft et al. testing ten blocks of 100% PLA printed on a single printer and Brown et al. testing similar objects outsourced to commercial printers. CT number variations of up to 121 and 226 HU were reported, respectively [[9], [35]]. The variation of 180 HU for the 100% infill blocks in this study fell between these values. This demonstrated the increased variation between different clinical systems, validating the need for inter-departmental audit, while indicating smaller errors than commercial prints. The mass density range for the blocks with 100% infill, 1.11–1.26 g/cm3, was consistent with published ranges for similar objects of 1.06–1.25 g/cm3 [[3], [9], [16], [18], [35], [36], [37], [38]]. The strong linear trend between inner densities and mean CT numbers (Fig. 4) was consistent with expectations. This confirmed that printed PLA exhibited predictable radiological behaviour and supported standard HU-density calibration for dosimetric modelling of 3D-printed components. The lack of a systematic shift associated with enhanced PLA suggests that filament formulation did not materially contribute to the inter-centre variability observed. Fig. 5(a) suggested a close relationship to the reference TPR values for water, reaffirming PLA's suitability as a tissue substitute [[20], [39]]. Fig. 5(b) showed that the calculated TPReff values from the attenuation measurements also fell within ± 2 SD of the mean values, indicating that reference curves for photon interactions with PLA may be feasible. The maximum inter-centre difference in measured TPR was 1.2% at 5 cm WET. This is below the ± 5% clinical dose tolerance recommended by the International Commission on Radiation Units and Measurements (ICRU) [40], and is therefore unlikely to produce clinically meaningful differences in delivered dose. WETs beyond 5 cm were not investigated and may produce larger inter‑centre variations. The similarity between the data measured using the participants' and reference CT images meant that participants could use locally scanned images for analysis, eliminating the need for a reference CT. This would allow centres to locally check their systems against the data presented here and participate in future audit activity remotely. Current advice is that centres should establish local relationships between infill and dosimetric properties [3], [18]. However, the mean %WE values presented here offer a starting point for the practical measurements required. There were some limitations to the audit process. Principally, the use of only three data points to determine the water equivalent data limited the accuracy of these calculations. Measurement of the 1 mm holes using digital callipers was subject to additional uncertainty from alignment limitations. Callipers provided a practical, reproducible approach, though higher-resolution methods such as optical microscopy could be explored in future work. Assumptions that prints accurately reproduced set outer wall thicknesses and printed these at 100% infill were not verified; however, any plausible deviations in wall thickness and density are expected to have minimal impact relative to the observed inter-centre variation. Non-cancellation from deviation in the printed hills and holes was also expected to be negligible. Changes in scatter conditions caused by different block supports used at imaging were assumed to have negligible effects on HU measurements, based on the use of radiolucent supports and the low-infill (20%) jig. This is supported by the absence of significant differences between %WE and ρWE derived from the reference and participant CT data, although minor variability cannot be excluded. Similarly, while scanner- and protocol-dependent HU variability exists [[41], [42]], all centres used 120 kV, controlling the parameter most associated with HU shifts. Modest HU differences produce small dosimetric effects in photon planning [43], consistent with our derived results. The multitude of available printer and software settings may create significant deviation from settings used in this audit, potentially without user awareness. If standardised data were to be generated, a mechanism to test individual system compatibility would be recommended. Future work will involve expanding the audit to a broader radiotherapy community and including a wider range of beam energies, to further investigate variations in the dosimetric properties of PLA across the therapeutic photon spectra. In conclusion, this multicentre audit characterised the geometric and dosimetric properties of FDM-printed PLA objects across clinical centres. Observed variability fell within ±2 SD of mean values, providing quantitative benchmarks to inform routine QA and support standardisation of 3D printing in radiotherapy. CRediT authorship contribution statement Matthew R. Jones: . Peter D. Woolliams: Resources. Matthew A. Bolt: Writing – review & editing, Resources. Owen McLaughlin: Resources. Maria Boutros: Resources. Tristan Wright: Resources. Rhys Jenkins: Resources. Anna Tonino: Resources. Fiona Milliken: Resources. Neil Bentley: Resources. Gordon D. Sands: Writing – review & editing. James C.L. Burnley: Writing – review & editing. Paul J. Doolan: Writing – review & editing. Dualta McQuaid: Writing – review & editing. Geoff J. Budgell: Writing – review & editing. Elizabeth J. Adams: Supervision. Conor K. McGarry: Writing – review & editing, Supervision, Resources. Catharine H. Clark: Writing – review & editing, Supervision, Resources. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.