An Adaptable Computed Tomography-Derived Three-Dimensional-Printed Alignment Fixture Minimizes Errors in Radius Biomechanical Testing

Benchtop cadaveric biomechanical testing is the gold standard for evaluating both the mechanical properties of bone and the performance of orthopedic implants. However, the bony geometries and mechanical properties of cadaveric specimens have inherent variability, which often limits the statistical power of findings and requires large sample sizes [1]. This issue can be readily circumvented with the use of synthetic bone models, but these surrogates are not always appropriate (e.g., modeling the mechanical behavior of cadaveric osteoporotic bone [1–3]). In these instances, cadaveric specimens still provide the best representation of the clinical experience, but best practices for minimizing variability in mechanical testing methods have been slow to emerge. In a recent literature review on experimental approaches to measure bone stiffness via compression testing, Zhao et al. found that all aspects of experimental protocols were variable, particularly the testing configuration [4].

Currently, there are no available standards for implementing repeatable specimen alignment protocols to determine the mechanical properties of long bones. The lack of a universally accepted approach is in part due to the difficulties associated with testing cadaveric bones. Challenges include the heterogeneity of specimen cohorts, residual soft tissue interfaces between the bone and test fixture, and in some instances, the low forces required to avoid specimen damage [4]. In addition, achieving consistent and accurate specimen alignment plays a key role in minimizing errors during biomechanical testing of cadavers [4–7].

Several animal studies have developed techniques and fixtures to ensure the repeatability and accuracy of specimen placement during biomechanical testing. For example, Bell et al. utilized a custom-designed breakaway mold to pot rat femora for mechanical testing of bone-implant anchorage [8]. This device centered the specimen and ensured that loading was directed perpendicular to the implant surface. Cheong et al. used microcomputed tomography (CT) scans of ovine tibiae in conjunction with a semi-automated algorithm to define principal components and subject-specific coordinate systems [9]. This allowed for pure bending and torsion during virtual mechanical testing; however, physical testing was never conducted, nor was the alignment of the virtually defined coordinate system to the machine's loading axis investigated.

To allow for direct comparisons between studies, researchers must conduct a clearly defined test through the use of a standard reference system and an alignment/potting fixture [10,11]. Unfortunately, many studies do not report the specimen alignment/potting protocol in sufficient detail for it to be replicated or assessed appropriately [12–14]. Proper alignment is difficult to achieve in practice because cross-sectional profiles of bones are highly variable and there are gradual curvatures along the length of the bone [8]. Typically, alignment is based on identification of anatomical landmarks that define local coordinate systems and reference axes [9,10,15,16]. Sample alignment is performed under visual inspection, but it is challenging to ensure that a specimen's theoretical reference axis is positioned collinear to the test frame mechanical axis and this detail is commonly overlooked [3,14,17,18]. Currently, there are no accepted standards for implementing controlled, repeatable whole-bone specimen alignment [4–7].

Accordingly, the objective of this study was to utilize medical imaging software and additive manufacturing techniques to design an adaptable, specimen-specific potting fixture capable of creating a repeatable alignment between bone and test frame axes. To achieve this goal, we developed an inexpensive and easy-to-replicate alignment/potting fixture that can be utilized on any long bone. To assess the quantitative impact of specimen malalignment on the results of long-bone mechanical tests, we compared physical and virtual mechanical tests. Our hypothesis was that specimen malalignment would be associated with error in the inferred axial stiffness and torsional rigidity of long bones.

Three matched pairs of fresh-frozen cadaveric radii were utilized in this study (one male and two females; age range 83–89 years). All donors had confirmed osteopenia by dual-energy X-ray absorptiometry scanning of the lumbar spine (t-scores: −1.1, −1.4, and −2.0). These specimens showed no evidence of any deformities, pathological abnormalities, or prior surgery.

Each forearm underwent clinical CT scanning in a Siemens Somatom Definition Edge scanner (Siemens Healthcare GmbH; Erlangen, Germany). An axial slice thickness of 0.5 mm, a pixel size of 0.625 mm, and a peak kilovoltage of 55 kVp were specified for each scan. A hydroxyapatite (HA) bone density calibration phantom (QRM-BDC/6; QRM GmbH) was included in each scan to convert radiodensity $ρHU$ (Hounsfield Unit, HU) to bone mineral density $ρQCT$ (quantitative computed tomography, QCT; mgHA/cm³).

Each CT scan was transferred for postprocessing using the digital imaging communications in medicine (DICOM) format. These DICOM files were imported into Mimics (v21.0; Materialize, Leuven, Belgium), which was used to transform the two-dimensional stacked image data sets into three-dimensional (3D) models (Figs. 1(a) and 1(b)). A predefined segmentation filter for bone (threshold range set to 226+ HU) was used to segment the hard tissue from the surrounding soft tissue (Fig. 1(b)). After the threshold-based segmentation, various other segmentation tools (slice edit, region grow, etc.) were used to isolate the radius in the scan (Fig. 1(c)). Each of the six 18-mm diameter cylindrical inserts in the phantom (0, 100, 200, 400, 600, and 800 mgHA/cm³) were also segmented for later calibration (see Sec. 2.2.4Material Assignment below).

In the body, the radius and ulna bear load together through the forearm. However, the purpose of this study was to develop a generalizable technique for aligning and mechanically testing any long bone, so the radius was used on its own. When examined in isolation, the mechanical axis of the radius is defined as the line that connects a point located on the center of the radial head to another point located in the middle of the interfossa ridge, between the scaphoid and lunate fossae on the distal radial articular surface, which is very similar to the International Society of Biomechanics proposed joint coordinate system for the radius [19–21]. To virtually realign each patient-specific radius model, first a spline was created in Mimics that connects the most distal point of the styloid process to the center of the ulnar notch [22]. Then a reference line was created, which connects the center of the radial head with the center of this spline. Reslicing the image stack using this reference line ensured that the proximal and distal ends of the radius are centered in the transverse plane and the diaphysis is vertical in the coronal and sagittal planes (Fig. 1(c)). This process was repeated for all the radii to ensure that they were similarly oriented.

As will be discussed in greater detail in Sec. 2.3Biomechanical Testing, the distal and proximal ends of each cadaveric radius were fixed in a urethane potting material to a depth equal to the widest portion of the distal radius. To replicate this condition in the virtual models, the distal and proximal ends of each radius model were cropped at the same levels. These surfaces were specified for boundary condition application.

Since the physical samples could only be potted once, deliberate malalignment was only introduced into the virtual models. In addition to the aligned configuration, in which the mechanical loading axis is aligned with the model's reference line, six other configurations were simulated: four off-centered configurations and two angular malalignment configuration. These models were configured to evaluate the effect of nonconcentricity (off-centered, but parallel) and angular malalignment between the mechanical axis of the bone and the test frame. The off-centered configuration was created by shifting the radius 10 mm anteriorly (A), posteriorly (P), medially (M), or laterally (L) relative to the mechanical loading axis (Fig. 2(a)). For the angular malalignment configuration, the distal surface was tilted 5 deg away from the reference line in the coronal (M-L angled) or sagittal plane (A-P angled) (Figs. 2(b) and 2(c)).

This material model was selected because the donors had documented osteopenia, which causes trabecularization of cortical bone [24,25]. These material assignment relationships, along with a Poisson's ratio of 0.3, were applied to all of the elements within each model using Mimics (Fig. 1(d)) [26].

All meshes were generated using 3-Matic (14.0; Materialize, Leuven, Belgium) and finite element analysis was executed using ANSYS Workbench Mechanical (ANSYS, Inc., Canonsburg, PA). In all models, the anatomic proximal cropped surface (elbow) was fixed. Physical tests were conducted with the radii in an inverted orientation, so in all models, the anatomic proximal surface was positioned below the anatomic distal surface (elbow below wrist). For the virtual torsion test, a 5-deg angular remote displacement about the axial reference line was applied to the anatomic distal cropped surface. A remote displacement was used to ensure that the angular displacement was applied about the reference axis (i.e., mechanical axis of the machine) and not the centroid of the anatomic distal cropped surface. For the virtual axial test, a compressive force of 250 N was applied at the anatomic distal cropped surface using a remote force boundary condition to ensure that the load was being applied through the reference axis. This load range was selected based on the forces that the articular radial surface experiences during light active motion [27–31]. The maximum displacement [mm] and moment reaction (N-mm) along the z-axis were recorded.

A mesh convergence study was performed to ensure that the key outcome measure used for comparison to physical torsion testing (calculated moment reaction) would not be affected by changing the size of the mesh. The mesh size was reduced until the result showed less than one percent difference in the moment reaction from the previous tested mesh size. As a result, a quadratic tetrahedral element type with a maximum edge length of 0.75 mm was used (Fig. 1(d)). The models had a median of 1.05 × 10⁶ elements.

Custom 3D-printed alignment/potting fixtures were built to ensure precise alignment of each radius with the mechanical axis of the test frame. The alignment fixture contained a modular two-piece insert printed with polylactic acid (Micromake Kossel Delta 3D, Shandong, China), where the cavity created by the assembled insert matched the mid-diaphyseal region of each specimen (Figs. 3(a)–3(c)). An outer universal sleeve was printed in Onyx (Mark II, Markforged Inc., MA) and similarly assembled to secure the insert (Fig. 3(d)). Once fully assembled, motion of the aligned bone was restricted to sliding on vertical stainless-steel rods, which were secured in an Onyx baseplate (Fig. 3(e)). All 3D rendered parts were designed in SolidWorks 2018 (Dassault Systemes SolidWorks Corps., MA). A detailed list of the 3D-printed parts and the fastening hardware can be found in the bill of materials in Digital Content I available in the Supplemental Materials on the ASME Digital Collection. A SolidWorks “Pack and Go” file containing the reference files needed to rebuild the assembly can be found in Digital Content II available in the Supplemental Materials on the ASME Digital Collection.

On the day of testing, specimens were thawed and all soft tissue was carefully removed. Specimens were wrapped in saline-soaked gauze, vacuum sealed, and refrigerated at 4 °C while awaiting testing. Freezing and thawing prior to mechanical testing has been shown to have a negligible effect on bone stiffness [4]. Anterior–posterior and medial-lateral radiographs were obtained (SireMobil, Siemens, Munich, Germany). Proximal and distal epiphyses of each specimen were potted in urethane (Master Dyna-Cast, Freeman Manufacturing and Supply, Mount Joy, PA) using the alignment/potting fixture as described in Sec. 2.3.1 Alignment/Potting Fixture. The potting depth was equal to the widest portion of the distal end, which was predetermined in the virtual environment for each bone. The depth was marked on the physical specimen using a set of digital calipers and an ink pen.

All specimens underwent the same nondestructive torsional mechanical testing protocol and specimen test order was randomized. Specimens were brought to room temperature (23 °C) before testing on a universal testing frame (Electro-Force 3550, TA Instruments, Eden Prairie, MN) equipped with a 1110 N/14.1 N·m load/torque cell. Two 3D-printed (Onyx) mounting cups supported the casting material on the proximal and distal ends of the radii. Next, the crosshead was lowered until the rim of both cups was flushed with each other. The bottom cup was then repositioned with an X-Y table such that the exterior walls of the cups were coplanar with one another. The crosshead was then raised and the potted specimens were securely bolted into the cups. The distal end of the radius was always fixed to the stationary portion of the test frame (Fig. 4). A universal joint was included in series with the specimen. Specimens were subjected to nondestructive cyclic loading under displacement control. Internal rotation to a maximum of 1.5 N·m was recorded at a rate of 1 deg/sec, representing a quasi-static loading condition, for 15 cycles. This torque value was not based on physiological loading, but instead it was chosen to impart a nondestructive torque for material characterization which is similar to other studies [8,9,15].

Specimen torsional stiffness, $ks$⁠, was then converted to torsional rigidity, $KT$⁠, by multiplying by the specimen gauge length: $KT=ksL$⁠.

where $F$ is the compressive load applied and $δ$ is the maximum total displacement in the model.

All statistical analyses were performed in spss (v.24; IBM Corp., Armonk, NY). Pearson's correlations were used to assess associations between variables. Deviations from normality were assessed using Shapiro-Wilk testing. Between-groups comparisons of aligned versus malaligned configurations in the finite element models were assessed using nonparametric Friedman tests. Post hoc pairwise comparisons were performed with a Bonferroni correction for multiple comparisons. Statistical significance was determined at an alpha of 0.05.

Initial verification of axis alignment between benchtop and virtual models was confirmed by superimposition of radiographs and model renderings (Fig. 5). Renderings of the virtual models, as viewed through the front and right planes in solidworks, were superimposed over the radiographs to align the global coordinate systems (z-axis on center with the alignment fixture). Without any manipulation of either image, the profile of the virtual models matched the profile of the corresponding radiograph of the clamped specimen, thus confirming that the fixture maintained the alignment that was created virtually. Excellent alignment registration was observed between the physical and virtual models in both views. For each donor, right-left radius pairs had similar torsional rigidities. The two female donors had a measured torsional rigidity 3.3 times lower than the male donor (specimens 2 and 3 versus 1; see Table 1).

The physical torsional rigidity values, $KT$⁠, compared favorably with the predicted virtual torsional rigidity values, $KVT$⁠, from the finite element models. A strong and statistically significant correlation between the physical torsional test results and virtual torsional test results was observed (R² = 0.9536, p < 0.001). Relative error between the physical and virtual torsional rigidities was highest for the male donor.

The mean and standard deviation results for virtual torsional rigidity in all configurations are shown in Table 2. For the virtual torsion test, the absolute percent error in the inferred torsional rigidity for the four off-centered configurations relative to the perfectly aligned configuration is shown in Fig. 6 for each specimen. In all donors, the largest percent errors occurred when specimens were positioned off-centered by 10 mm in the anterior and medial directions, but these errors were less than 4% in all cases. The mean and standard deviation of the absolute percent errors in inferred rigidity in the four off-centered configurations (anterior, posterior, medial, and lateral) were 2.2% ± 2.5%, 0.7% ± 0.4%, 1.8% ± 2.2%, and 0.2% ± 0.3%, respectively. Overall, there were significant differences in $KVT$ associated with alignment, $X2$(4) = 10.953, p = 0.027, but post hoc analysis revealed no significant individual pairwise comparisons between alignment groups after Bonferroni correction of the p values.

The absolute percent error in the inferred torsional rigidity for the angular malaligned cases relative to the aligned configuration is shown in Fig. 7(a) for each specimen. Errors were similar for rotations occurring in the coronal and sagittal planes. The mean and standard deviation of the absolute percent errors in inferred rigidity in the two angular malalignment configurations (coronal and sagittal) were 6.2% ± 2.7% and 8.0% ± 3.2%, respectively. Overall, there were significant differences in $KVT$ associated with alignment, $X2$(2) = 10.333, p = 0.006. Posthoc analysis revealed a significant individual pairwise comparison between the sagittal and aligned groups (adjusted significance p = 0.004).

The mean and standard deviation results for virtual axial stiffness in all configurations are shown in Table 2. For the simulated axial compression test, the absolute percent error in the inferred stiffness for the four off-centered configurations relative to the perfectly aligned configuration is shown in Fig. 6 for each specimen. Overall, the malalignment-induced percentage errors were an order of magnitude higher in axial compression testing than in torsional testing. As before, in all donors, the largest percent errors occurred when specimens were positioned off-centered in the anterior and medial directions. The mean and standard deviation of the absolute percent errors in inferred stiffness in the four off-centered configurations (anterior, posterior, medial, and lateral) were 32.5% ± 5.3%, 3.5% ±17.6%, 40.2% ± 6.4%, and 9.4% ± 15.2% , respectively. Overall, there were significant differences in axial stiffness associated with alignment, $X2$(4) = 20.800, p < 0.001. Posthoc analysis revealed multiple significant individual pairwise comparisons, including medial-posterior, medial-aligned, medial-lateral, and anterior-lateral (adjusted significances p ≤ 0.035).

The absolute percent error in the inferred axial stiffness for the angular malalignment cases relative to the aligned configuration is shown in Fig. 7(b) for each specimen. Errors were similar for rotations occurring in the coronal and sagittal planes. The mean and standard deviation of the absolute percent errors in inferred stiffness in the two angular malalignment configurations (coronal and sagittal) were 129.1% ± 24.7% and 127.4% ± 17.5%, respectively. Overall, there were significant differences in axial stiffness associated with alignment, $X2$(2) = 9.000, p = 0.011. Posthoc analysis revealed two significant individual pairwise comparisons between both angular malalignment cases and the aligned group (adjusted significances p ≤ 0.028).

This study demonstrates the utility of an adaptable CT-derived, 3D-printed alignment fixture that effectively minimizes errors in whole-bone biomechanical testing. Previous applications have included anatomy-specific fixtures to assist with positioning of specimens prior to mechanical testing, pre-operative planning, and development of patient-specific implants [32–36]. Though these studies demonstrate how 3D printing technology can be utilized to create more robust test setups, widespread adoption has been lagging. This study provides an open-source, simple solution to this problem for whole-bone testing applications, specifically the need to combine a consistent anatomic reference system with the use of a robust alignment protocol and potting fixture. This fixture can be adapted for use with any long bone and uses a customized 3D-printed insert to align each specimen precisely to a reference line that is chosen a priori in a virtual modeling environment. The only part that is specimen-specific is the cavity insert that functions as a negative mold of the specimen; every other component can be reused. The insert can be edited using a multitude of free 3D modeling software options (e.g., Auto-Desk Fusion 360, among others). Barriers to entry include the necessity for CT scanning or 3D surface scanning of the specimen before the test, access to software for image segmentation and/or surface modeling, and access to 3D printing resources. These facilities and resources are now becoming more widely available in many academic institutions. The extra time and cost incurred to develop the fixture in this study is relatively small compared to the cost of cadaveric specimens and orthopedic implants. For research teams without institutional access to 3D printing, many affordable, web-based 3D printing services (e.g., Shapeways, 3D Hubs, Scuplteo, etc.) are also available, which allows for low-cost fabrication of the parts and alleviates the need to invest in a 3D printer.

The results from this study confirmed our hypothesis that specimen malalignment is associated with error in biomechanical testing. In particular, we found that axial stiffness measures were very sensitive to malalignments between the specimen mechanical axis and the test frame axis, with average absolute errors up to 129%. Errors were highest when the offset produced bending in the same direction as the natural curvature of the radius. In contrast, torsional rigidity measures were more robust to deviations from ideal alignment, with errors <4% due to off-center malalignment and <15% due to angular malalignment. Our findings are consistent with previous reports that have emphasized the importance of achieving repeatable alignment of test samples in biomechanical testing. Cristofolini and Viceconti (2000) applied the anatomic reference system proposed by Ruff and Hayes (1983) to align cadaveric and composite tibiae for mechanical testing, but found that achieving ideal alignment was difficult in practice, with significant potential for error particularly in axial loading [3,16]. Even when each specimen was marked by a robust and consistent reference system, ensuring that this reference system was aligned with the test frame loading axis proved difficult. This suggests that whenever appropriate, torsion tests should be used preferentially as a summary mechanical measure, or as a supplement to axial compressive tests.

Limitations of this study include the small sample size of cadaveric specimens (only three donors) and limited available physical test data (torsion only) for comparison between physical and virtual tests. The virtual models are further limited by the use of a material assignment law that is not specific to mixed cortical/cancellous bone. Numerous scaling relations for radiodensity and Young's modulus are available in the literature for different species, anatomic sites, and bone types [37]. We chose Eq. (2) after evaluating the performance of three candidate scaling equations, one for human cortical bone, one for human trabecular bone, and one for nonosteoporotic human radii (Digital Content III available in the Supplemental Materials on the ASME Digital Collection) [23,38,39]. The nonosteoporotic human radii law overpredicted the rigidity of our radii, which we attribute to the structural and mechanical changes known to occur in osteopenic cortical bone [38]. Conversely, the cortical bone law underpredicted the rigidity for these low density, osteopenic bones [39]. The pooled trabecular law we ultimately chose, although not derived for osteopenic radii, showed very close correspondence between physical and virtual tests for the two female donors [23]. The larger error between the physical and virtual tests for the male donor was likely due to the less-advanced progression of osteopenic bone degeneration, leading to an underprediction of rigidity with the trabecular material assignment law.

Specimen malalignment can cause significant errors in biomechanical testing of whole bones. The results from this study highlight the importance of combining a consistent anatomic reference system with the use of a robust alignment protocol and potting fixture. Torsion testing is more robust to alignment artifacts than axial compressive testing, but all loading modes can be improved by ensuring the desired alignment between anatomic reference lines and the test frame. To achieve this optimal alignment, we have designed a customizable specimen alignment and potting fixture that guarantees the desired sample alignment through image-based pretest planning and 3D printing. Use of this method may be particularly important for more sensitive biomechanical tests (e.g., axial compressive tests) that may be needed for industrial applications, such as implant design and construct mechanical testing.

The authors acknowledge Carolyn Persinger and Jaclyn Rissling for their support in obtaining and processing the clinical CT scans and Sarah Kauzmann for assistance with graphical design in Fig. 4.

• National Science Foundation (Grant No. CMMI-1943287; Funder ID: 10.13039/100000001).

CT =

computed tomography

DICOM =

digital imaging communications in medicine

E =

Young's modulus

F =

axial compressive load

HA =

hydroxyapatite

HU =

Hounsfield unit

$k =$ =

total torsional stiffness

$kf$ =

fixture torsional stiffness

$ks$ =

specimen torsional stiffness

$KT$ =

torsional rigidity

$kVA$ =

axial stiffness

$KVT$ =

virtual torsional rigidity

L =

specimen gauge length

M =

z-direction moment reaction

QCT =

quantitative computed tomography

$δ$ =

maximum total displacement

$ϕ$ =

angle of twist