The 3-D geometry of bones and joints can be determined from MRI or CT images, and MRI-based 3-D models of articular surfaces have been reported.1,2 Geometric accuracy of CT images have been described in several studies, including studies that focused on cortical thickness3 and geometry of the articular surface.4–7 Accuracy of CT, cone-beam CT, and micro-CT is influenced by several factors, including beam orientation relative to the scan surface and slice thickness.3 An increase in the popularity of additive manufacturing has led to the common use of CT data to create 3-D or physical models. Accuracy of additive manufacturing models produced from CT-derived 3-D models has been evaluated with a focus on CT scan variables and CT scan segmentation variables.8,9 Imaging with CT has been used to assess the shape and relative position of bones,10 estimate joint loads,11–13 conduct computational modeling of articular contact,14,15 assess pathological changes in joints,16,17 and plan surgical procedures.18
In addition to evaluation by use of CT, the 3-D geometry of articular surfaces has been evaluated by use of several methods that rely on ionizing radiation, line-of-sight scanning, or coordinate measurements. Stereophotogrammetry has been used to evaluate the shape of cartilage and subchondral bone.19 Laser coordinate measuring systems have also been used to evaluate bone shape20 and symmetry.21 Studies conducted to evaluate accuracy and validation of CT have involved the use of coordinate measuring systems as the criterion-referenced standard for the knee joints,22 phalanges,23 and vertebral column24 of humans.
Our research group has focused on designing limb-sparing implants for companion animals and designing custom implants intended to replace the articulating surfaces of small joints through custom hemiarthroplasty or custom total joint replacement. Little is known about the negative consequences of geometric inaccuracy of custom implants. It appears likely that geometric inaccuracy would have a negative impact for an implant articulating with another implant (ie, custom total joint arthroplasty) and an implant articulating with native cartilage (ie, custom hemiarthroplasty).
The carpus is a common site of limb-sparing procedures in companion animals. Therefore, the study reported here was conducted on the carpus of cats. We intended to evaluate whether obtaining CT images of a limb with the beam in a sagittal plane would increase geometric accuracy of the articular surface, compared with results for CT images obtained in a longitudinal orientation, considering that there would be a larger number of slices across the articular surface. We hypothesized that measurements for CT images obtained in a transverse orientation would be more accurate than measurements for CT images obtained in a longitudinal orientation. We also intended to compare accuracy of these 2 CT protocols with accuracy for micro-CT of a segment of the entire forelimb that simulated a clinical CT examination with extremely thin slices and with results for laser-scanned articular surfaces after cartilage digestion. We hypothesized that micro-CT measurements would have acceptable accuracy, whereas measurements obtained in longitudinal and sagittal orientations with CT would not have acceptable accuracy. Because our research group was interested in implant design, the focus of the study reported here was to compare geometry of the subchondral bone rather than geometry of the articular surface.
Materials and Methods
Sample
Forelimbs of 12 cats euthanized at 3 local animal shelters for reasons unrelated to the study reported here were harvested and used. Skeletally immature cats (as determined on the basis of results of physical examination or CT) were excluded. Each forelimb was removed in its entirety at the shoulder joint with intact soft tissues. Forelimbs were wrapped in gauze soaked in saline (0.9% NaCl) solution, placed in an extended position in airtight containers, and frozen at −20°C.
An a priori power analysis conducted by use of results for a preliminary experiment performed with 5 specimens revealed that 24 specimens would be needed for the study reported here. Of the 24 specimens harvested, 1 was removed because of skeletal immaturity that was detected after cartilage digestion; thus, the study involved 23 radii. Size of the effects detected in the study were larger than values used for the a priori power analysis; therefore, removal of 1 specimen did not yield a consequential loss in power.
CT images
Forelimbs were thawed at 4°C overnight and then at room temperature (approx 22°C) for 2 hours. Images of thawed forelimbs were obtained by use of a multislice CT scannera with the following settings: 140 mA; 120 kVp; slice thickness, 0.6 mm; slice increment, 0.4 mm; 512 × 512 pixels; and voxel size, 0.6 mm. Images of the entire forelimb were obtained in 2 orientations: longitudinal orientation (CTLO images) with cross-sectional CT data collected in the transverse plane, and transverse orientation (CTTO images) with the cross-sectional CT data collected in the sagittal plane. The transverse orientation was used to maximize the number of slices for articular surface data.
Micro-CT images
A band saw was used to prepare a 4-cm-long section centered over the radiocarpal joint. Specimens were scanned by use of a micro-CT scannerb with the following settings: 150 kV; 14.6 W; and voxel size, 0.024 mm.
Laser scans
Enzymatic digestion was used to remove the articular cartilage of the distal portion of the radius without disturbing the subchondral bone.25 The optimal digestion period was determined in the preliminary experiment. In that experiment, 5 radii of cats were immersed in a papain solution (papain [10 mg/mL], 5.5mM l-cysteine, and 10% penicillin-streptomycin-amphotericin in PBS solution) at 37°C. The bones were dried and weighed at 30 minutes and 2, 6, 12, 18, 24, 36, 48, 60, and 72 hours. There was a negligible change in weight of bones between 36 and 72 hours of digestion. Results of the preliminary experiment indicated that cartilage was removed by 36 hours and that no appreciable bone loss occurred during the 72 hours. Therefore, a 36-hour digestion period was selected.
Carpal joints of the 24 specimens were dissected, and the articular cartilage of the distal portion of the radius was enzymatically digested. Specimens were rinsed with PBS solution, lightly coated with talcum powder, and scanned by use of a laser scannerc with an accuracy of 0.025 mm to acquire point clouds of the articular surface of the distal portion of the radius.
Reconstruction and analysis methods
The CTLO, CTTO, and micro-CT images were imported as DICOM images by use of image processing software.d The DICOM images were used to reconstruct 3-D models of the radius by use of a marching cubes algorithm. The CT images were filtered and selected to include grayscale values from 226 to 3,053 HU (standard thresholds for bone segmentation). When image manipulation was required to separate the radius from the radial carpal bone and ulna, geometry of the articular surface of the radius was not altered. Micro-CT images were filtered and selected to include all grayscale values > 2,358 HU. Threshold used for the segmentation of micro-CT images was determined from micro-CT images of the 5 radii obtained during the preliminary experiment and that were segmented at various thresholds. The threshold deemed most accurate for the preliminary study (2,358 HU) was selected for use in the study reported here. Additional steps were not required to isolate the radius from the radial carpal bone and ulna. Output files (.stl format) were generated by use of an optimal generation setting to avoid smoothing or data averaging during processing.
Laser scans were registered to create a single point cloud model for each specimen by use of automatic registration in the modeling software.e Outliers and disconnected components were removed prior to creation of .stl files of the laser-scanned bone models; the .stl files of the laser-scanned 3-D point cloud bone models were used as the criterion-referenced standard. The .stl files for CTLO, CTTO, and micro-CT images were compared with the .stl files of the laser-scanned 3-D model point cloud data. Anatomic reference planes (coronal, parasagittal, and transverse) were assigned to each model of the laser scan data by use of computer-aided design software.f Bounds of the radial articular surface were determined manually and separated in the computer-aided design software, and coronal, transverse, and parasagittal anatomic planes were aligned with a global coordinate system. The .stl files were imported by use of a high-level programming languageg into the modeling software.e A 2-phase automated best fit alignment of the .stl files of the laser scan, CTLO, CTTO, and micro-CT images was performed. An initial alignment was performed with high-precision fitting and a tolerance of 1 × 10−6 mm. A second alignment was performed with high-precision fitting and fine adjustments to further reduce fitting errors. To evaluate the craniocaudal radius of curvature, planes that were parallel to the parasagittal plane and spaced 0.5 mm apart were created across the articular surface of the laser scan model by use of an automation feature; identical planes were created for the 3-D models of the CTLO, CTTO, and micro-CT images. The radius of curvature was calculated by use of a 2-D measurement feature, with a rectangular box used to determine the bounds of the articular surface and the radius calculation performed by use of the best fit option. Investigators created the rectangular box used to determine the bounds of the articular surface on the basis of the curved region of interest in the criterion-referenced standard of the laser-scan models and duplicated it for all other images by use of the automation feature. Similarly, mediolateral radius of curvature was calculated within planes that were parallel to the coronal anatomic plane and spaced 0.5 mm apart.
Calculations of surface deviations were performed by use of the quality control softwaree in accordance with the same 2-phase best fit alignment described previously. The articular surface of the laser scan point cloud was manually selected, and that region was compared among images by use of a 3-D deviation tool (maximum deviation, 10 mm; maximum angle, 45°) to calculate mean and maximum positive and negative deviations.
The craniocaudal and mediolateral lengths were measured by use of .stl files. The .stl files were imported in the computer-aided design softwaref and aligned by use of an initial manual n-points alignment (3 points on the articular surface and 2 points on the bone shaft), wherein the models for the CTLO, CTTO, and micro-CT images were compared with the model for the laser scan. An additional global alignment was performed by use of an automatic alignment feature (tolerance, 1 × 10−6 mm). Planes parallel to the craniocaudal plane were created at the most dorsal and most palmar aspects of the distal portion of the radius.26 Planes parallel to the mediolateral plane were created at the most medial and most lateral aspects of the distal portion of the radius. Distance between the most dorsal and most palmar plane was recorded as craniocaudal length, and distance between the most medial and most lateral plane was recorded as the mediolateral length.
Statistical analysis
Accuracy thresholds were selected by extrapolating information from studies27,28 in which consequences of implant mismatches or articular cartilage steps were described. Accuracy was considered acceptable when the magnitude of the deviation was < 0.6 mm (slice thickness selected for use in the study) for linear dimensions, < 0.75 mm for the radius of curvature,27 and < 50 μm for mean deviations from the articular surface.
A complete block design statistical model was developed by use of the differences between the images for the various modalities and the laser scan. Residual effects were assumed to have a normal distribution with a mean of 0 and nonequal variance with respect to treatment. A generalized linear mixed model was used to analyze covariance.h Values were considered significantly different at P < 0.05.
Results
The mean ± SD percentage error was larger for CTLO and CTTO images than micro-CT images for all variables measured (Table 1). Mean mediolateral lengths of CTLO, CTTO, and micro-CT images differed from measurements for the laser scan; however, mean mediolateral lengths did not differ significantly among CTLO, CTTO, and micro-CT images. Mean craniocaudal length differed among CTLO, CTTO, and micro-CT images. Mean error for the mediolateral radius of curvature was larger for CTLO and CTTO images than for micro-CT images (Figures 1 and 2). Mean error for the craniocaudal radius of curvature did not differ significantly among CTLO, CTTO, and micro-CT images. Root mean square error of the deviation from the articular surface was larger for models of CTLO and CTTO images than for models of micro-CT images (Figure 3). Mean maximum positive surface deviation was larger for CTLO and CTTO images than for micro-CT images. Mean maximum negative surface deviation differed significantly among all CT images; it was smallest for micro-CT images and largest for CTTO images. Variance for micro-CT images was significantly less than variance for CTLO and CTTO images for all variables measured. Significant differences in variance between CTLO and CTTO images were detected for only the craniocaudal radius of curvature, wherein the variance for CTTO images was larger than the variance for CTLO images.
Box-and-whisker plots of differences between measurements derived from CT images obtained in longitudinal orientation (CTLO images) with cross-sectional CT data collected in the transverse plane, CT images obtained in transverse orientation (CTTO images) with cross-sectional CT data collected in the sagittal plane, and micro-CT images and coordinate measurement machine measurements of the mediolateral (ML) and craniocaudal (CC) radius of curvature of the articular surface of the distal portion of the radius of 23 specimens obtained from feline cadavers. All modalities underestimated the ML radius of curvature. The boxes represent the second and third quartiles, the horizontal line in each box represents the median, whiskers represent the highest and lowest values, and circles represent outliers.
Citation: American Journal of Veterinary Research 80, 10; 10.2460/ajvr.80.10.976
Representative mediolateral cross-sectional measurements of articular joint surfaces of the distal portion of the radius of feline cadavers comparing results for CTLO images (dashed line; A), CTTO images (dotted line; B), and micro-CT images (dashed-and-dotted line; C) with results for a coordinate measurement machine (solid lines). Notice that for CTLO and CTTO images there are large positive and negative deviations from the results for the coordinate measurement machine. In comparison, results for micro-CT images closely approximate results for the coordinate measurement machine. Bar = 1 mm. See Figure 1 for remainder of key.
Citation: American Journal of Veterinary Research 80, 10; 10.2460/ajvr.80.10.976
Representative 3-D models (A, B, D, F) and heat maps of deviations from the articular surface (C, E, G) for the radius of a feline cadaver. Positive differences between the models created by use of laser-scanned images (A) and CTLO images (blue areas; B), CTTO images (purple areas; D), and micro-CT images (pink areas; F) are indicated. For the heat maps, negative deviations (light to dark blue) indicate that the CT models overestimated the articular surface, and positive deviations (yellow, orange, or red) indicate that the CT models underestimated the articular surface. Notice that the scale of the heat map differs among panels. Deviations are approximately 10 times as large for the CTLO and CTTO images as for the micro-CT images. Bar = 5 mm.
Citation: American Journal of Veterinary Research 80, 10; 10.2460/ajvr.80.10.976
Mean ± SD values for measurements obtained from CT, micro-CT, and laser scan images of the articular surface of the distal portion of the radius (n = 23) of 12 feline cadavers*
| Variable | Laser scan images | CTLO images | CTTO images | Micro-CT images |
|---|---|---|---|---|
| ML length | ||||
| Length (mm) | 12.64 ± 0.92 | 13.06 ± 0.84 | 13.13 ± 0.88 | 12.47 ± 0.81 |
| Difference from laser scan (mm) | NA | 0.43a ± 0.41d | 0.49a ± 0.41d | –0.17b ± 0.16e |
| Error (%) | NA | 3.62 ± 3.31 | 4.48 ± 2.59 | 1.30 ± 1.12 |
| CC length | ||||
| Length (mm) | 8.37 ± 0.66 | 8.86 ± 0.67 | 9.20 ± 0.72 | 8.34 ± 0.62 |
| Difference from laser scan (mm) | NA | 0.49a ± 0.20d | 0.83b ± 0.26d | –0.03c ± 0.10e |
| Error (%) | NA | 5.91 ± 2.37 | 9.97 ± 3.08 | 0.82 ± 0.86 |
| ML radius of curvature | ||||
| Difference from laser scan (mm) | NA | –0.37a ± 0.93d | –0.31a ± 1.15d | –0.19b ± 0.24e |
| Error (%) | NA | 12.04 ± 10.48 | 14.25 ± 18.06 | 2.52 ± 2.79 |
| CC radius of curvature | ||||
| Difference from laser scan (mm) | NA | 2.58a ± 22.55d | 5.48a ± 41.99e | –0.10b ± 0.27f |
| Error (%) | NA | 99.41 ± 549.31 | 256.57 ± 1,716.59 | 5.18 ± 4.80 |
| Surface deviation | ||||
| RMSE (mm) | NA | 0.26a ± 0.09d | 0.30a ± 0.28e | 0.04b ± 0.02d |
| Maximum positive deviation (mm) | NA | 3.04a ± 1.17d | 2.45a ± 0.90d | 0.38b ± 0.18e |
| Maximum negative deviation (mm) | NA | –1.18a ± 1.11d | –1.53b ± 0.73d | –0.26c ± 0.11e |
The CT images were obtained in longitudinal orientation (CTLO images) with cross-sectional CT data collected in the transverse plane and in transverse orientation (CTTO images) with cross-sectional CT data collected in the sagittal plane; laser scan images were evaluated with a coordinate measurement machine.
CC = Craniocaudal. ML = Mediolateral. NA = Not applicable. RMSE = Root mean square error.
Within a row, mean values with different superscript letters differ significantly (P < 0.05).
Within a row, SD values with different superscript letters differ significantly (P < 0.05).
Discussion
For the study reported here, geometry of 3-D models derived from CTLO, CTTO, and micro-CT images were compared with each other and with models derived by use of a laser scanner to assess accuracy. The articular surface of the distal portion of the radius of cats was selected for evaluation because of its small size and curvature, because cadavers of cats euthanatized for reasons unrelated to this study were available for use, and because we anticipated that the geometry of the articular surface of the distal portion of the radius of cats would be more homogeneous than would the articular surface of the distal portion of the radius of dogs, which would require a smaller sample to detect differences in accuracy among CT protocols.
We selected the surface of a laser-scanned image as the criterion-referenced standard. Micro-CT images could have been selected as the criterion-referenced standard. However, 3-D models prepared by use of micro-CT images require image segmentation. Use of thresholds during segmentation influences accuracy of models.9 However, the preparation of 3-D models from laser-scanned surfaces did not require image segmentation and therefore was independent from scanning parameters and image thresholds. This would be particularly relevant for metaphyseal and subchondral bone, which are locations where cortical bone is thin. Had we opted to use micro-CT images as the criterion-referenced standard for the present study, the micro-CT protocol could have been enhanced by dissection of the radius before micro-CT images were obtained because the bone-air interface would enhance contrast of micro-CT images, compared with contrast for bone interfacing with joint fluid and soft tissues. However, the micro-CT protocol used in the study resembled a protocol used in clinical settings wherein a segment of the entire limb was scanned. As anticipated, micro-CT was more accurate than conventional CT. However, the use of micro-CT for clinical patients is not feasible because of high amounts of ionizing radiation and slow data acquisition. Other CT methods, including conventional CT with a voxel size < 0.6 mm and cone beam CT, would have provided greater accuracy but were not available in our teaching hospital. High-resolution cone beam CT images have slices as thin as 0.1 mm combined with low radiation, relative to slice thickness and radiation for conventional CT.29,30
The distal portion of the radius is a common site of limb-sparing surgery in companion animals31 and a site of hemiarthroplasty in humans.32 Accurate assessment of the geometry of joint surfaces is a prerequisite for optimization of articulating custom orthopedic implants (eg, custom arthroplasties or hemiarthroplasties). Accurate fit of articulating surfaces would also require accurate surgical placement of a custom implant. Angular variability and positional variability in joint replacement is improved with surgical navigation implantation33 and the use of patient-specific alignment instruments.34 Deviations in measurements of the articular surface (eg, deviations in length or the articular surface) differed among modalities. Accuracy and precision were lower for measurements obtained by use of CTLO and CTTO images, compared with measurements obtained by use of micro-CT images, and most of these differences (4/5 for mean and 5/5 for variance) were significantly different. We accepted the hypothesis that measurements for micro-CT images had acceptable accuracy and that measurements for CTLO and CTTO images did not have acceptable accuracy. The main source of inaccuracy for CTLO and CTTO images was likely a partial volume effect combined with spillover from the adjacent radial carpal bone. The fact that the bone shape was overestimated on CTLO and CTTO images suggested that spillover from the radial carpal bone was most influential. For the micro-CT images, the slight underestimation of dimensions was likely a consequence of the partial volume effect.
The lack of significant differences in mediolateral length among CTLO, CTTO, and micro-CT images may have resulted from the fact that mediolateral length had the largest linear measurements and imaging errors were smallest relative to overall size or to the relatively large variance for measurements of CTLO and CTTO images. Differences in the accuracy for micro-CT images and CTLO or CTTO images were likely a consequence of the increased resolution for the micro-CT images relative to that for the CTLO and CTTO images. Edges of voxels for CTLO and CTTO images were approximately 25 times as large as voxel edges for micro-CT images. Micro-CT images included approximately 15,000 more voxels than CTLO or CTTO images. Dimensions of a voxel comprise in-plane pixel size (a combination of the field of view and matrix dimensions) and slice thickness. A decrease in voxel volume increases geometric accuracy, but it would increase the radiation dose. In a study35 conducted to evaluate the influence of slice thickness on radiation doses of humans, a decrease in slice thickness increased the radiation dose.
Accuracy and precision of measurements did not differ significantly between CTLO and CTTO images for most linear measures of deviation (3/5 for mean and 5/5 for variance). Therefore, we rejected the hypothesis that use of CTTO images was more accurate than use of CTLO images. Lack of improvement in models created from CTTO images, as compared with CTLO images, was unexpected because there were > 3 times as many slices containing articular surface data for the CTTO images than for the CTLO images. Subjectively, 3-D models reconstructed from the CTTO images had more noise than the CTLO images, likely because of the increased length for x-ray travel within specimens. Beam hardening that appeared to be generated in the region of the elbow joint and radiated along the shaft of the radius to the distal portion of the radius may have been responsible for the lack of an increase in accuracy for CTTO images relative to CTLO images. Beam hardening has been described in the literature.36 Effects are more pronounced in bone because of the high calcium content37 and subsequent absorption of x-rays.38 Attempts to optimize corrections for beam hardening exist in clinical applications37 and in industrial metrological applications.38–40 Investigators of another study41 compared results for CTLO, oblique CT, and CTTO images of a human ankle prosthesis implanted in a stifle joint of a pig. In that study,41 measurements for CTTO images were less accurate than measurements for CTLO images.
Measurement errors, including errors in the radius of curvature, for models created by use of micro-CT images were small and compatible with the design of conforming implants. By comparison, measurement errors associated with CTLO and CTTO images were large, particularly errors in the radius of curvature measurements. The mediolateral radius of curvature was more accurately evaluated than was the craniocaudal radius of curvature. Additionally, the large percentage errors indicated that although there was no measurement bias (mean differences were centered around zero), the magnitude of error was large, which meant measurements were less reliable. The current resolution for CTLO and CTTO images would not be compatible with the design of custom-conforming arthroplasties or the evaluation of small topographic differences in small articular surfaces. Several studies have been conducted to evaluate the geometric aspects of the accuracy for CT of bones from other species. In a study42 conducted to compare the accuracy of models of human tibiae created by use of CT images (pixel size, 0.39 × 0.39 mm; slice thickness, 0.6 mm) and MRI scans by use of optical laser scanning, mean of the root mean square error for the entire bone was 0.55 mm, and mean of the root mean square error for the proximal and distal 10% of the bone was 0.64 and 0.65 mm, respectively. These errors are approximately twice as large as the root mean square error for CTLO and CTTO images in the study reported here. Interestingly, an increase in bone size (human tibia vs cat radius) and a decrease in x-y pixel size did not improve accuracy. In 1 study43 in which a calibration device with known dimensions was scanned by use of a slice thickness of 1 mm, mean geometric error was 0.17 mm. In another study23 in which a voxel size of 0.39 × 0.39 × 0.4 mm was used for CT, linear deviation in accuracy for bones of the index finger was 0.21 mm in the distal phalanx, 0.20 mm in the middle phalanx, and 0.19 mm in the proximal phalanx. These error values are similar to the root mean square error values obtained in the study reported here. This likely can be attributed to the fact that human phalanges are similar in size to feline radii. A study24 conducted to compare the accuracy of length measurements between anatomic landmarks of the vertebral column by use of a CT slice thickness of 1 mm resulted in mean ± SD global error of 1.1 ± 0.8 mm and errors in length measurements ranging from 0.8 to 2.4 mm. These errors are larger than the errors reported for the radii of the present study.
The variables used for CT influence CT accuracy. In a study8 conducted to evaluate the accuracy of replicas of a canine femur, 3 models were additively manufactured from CT data collected by use of various scan variables, and linear dimensions were compared by use of a touch coordinate measuring system. Various values for penetrability (80 to 140 kVp), tube current (60 to 160 mA), and pitch (0.6 or 0.75), and several retroreconstruction kernels and windows were used. The authors reported that 2% to 29% of the variability was attributable to differences in the model fabrication methods, and 4% to 44% of the variability was attributable to differences in CT scan variables. However, no direct measurements of the 3-D models before manufacturing were collected in that study.8 In the aforementioned study23 conducted to evaluate CT accuracy of human phalanges, voxel size was constant (0.12 × 0.12 × 1 mm) but penetrability (80 to 140 kVp) and tube current (80 or 100 mA) varied. However, segmentation thresholds and triangulation resolutions also varied, which complicated the assessment of the influence of CT scan variables on CT accuracy. In that study,23 investigators compared CT images of the human phalanges to CT images of the additive manufacturing models of the phalanges; however, no comparison was made directly with the bone model. Higher penetrability values resulted in a decrease in the number of voxels in the images of the additive manufacturing models as compared with images of the bones, and an increase in tube current resulted in an increase in voxels in the phalanges but not in the additive manufacturing models. Increases in the lower grayscale threshold resulted in a reduction in the number of voxels in the additive manufacturing model but not in the phalanges and increased the number of voxels in the phalanges but not in the additive manufacturing models. Decreasing the lower threshold resulted in an increase in the number of voxels in the phalanges but not in the additive manufacturing model.44 These changes illustrate the large influence that CT variables and segmentation have on the accuracy of CT data.
Investigators have used segmentation based on grayscale value filtering.8,9 Alternative segmentation strategies, including Canney edge detection and intensity thresholding, have been described. A study44 was conducted to compare CT images (pixel size, 0.39 × 0.39 mm; slice thickness, 0.5 mm) and 3-D contact scans of ovine femurs. Mean deviation was 0.24 mm for single threshold errors and 0.18 mm for multithreshold errors. Deviations in that study44 were similar to deviations in the study reported here. On the basis of the variety of scanning variables and segmentation methods available for use, acceptable scanning variables and segmentation methods for modeling of small joint surfaces will likely require comprehensive standardization of CT imaging protocols, retroreconstruction methods, and segmentation to optimize CT accuracy for specific regions of interest.
Optimizing CT accuracy is important when patient-specific computer-aided design data are used in the design of hemiarthroplasty implants. Increasing the conformity of a hemiarthroplasty can reduce friction and improve wear performance, as compared with in vivo wear.45 In hip joint hemiarthroplasty, undersize of the radius of a prosthetic head by as little as 0.75 mm (6.25%) negatively impacts the performance of the hemiarthroplasty.27 The large difference in maximum positive deviation and negative deviation and in radius of curvature among modalities in the present study warrants further investigation into the optimal data collection methods that can be used to accurately record the articular surface of feline radio-carpal joints and other small joints. Currently, evaluations conducted to assess the effect of articular surface geometry on cartilage contact stresses relies on finite-element modeling.14,15
Clinical CT imaging and grayscale filtering segmentation methods used to develop 3-D models in the study reported here did not have sufficient geometric accuracy for modeling small articular surfaces for the purpose of designing a surface that would conform to an articular surface, such as the surface of a prosthesis intended for hemiarthroplasty. Lack of accuracy of clinical CT imaging raises concerns about the potential lack of accuracy when planning the correction of a limb deformity or repair of an articular fracture. Streamlined protocols for imaging, segmentation, and smoothing should be developed to optimize CT accuracy.
Acknowledgments
This manuscript represents a portion of the dissertation submitted by Dr. Webster to the Edward P. Fitts Department of Industrial and Systems Engineering as partial fulfillment of the requirements for a Doctor of Philosophy degree.
Supported by the Leonard X. Bosack & Betty M. Kruger Charitable Foundation.
The authors declare that there were no conflicts of interest.
The authors thank Dr. Susan Bernacki, Dr. Ian Robertson, Dr. Harvey West II, James Robey, Jimmy Thostenson, Amanda Hanley, Diana Courtright, and Kristen Karasiewicz for technical assistance.
Footnotes
Siemens Sensation 64, Siemens, Washington, DC.
XT H 225 ST, Nikon, Melville, NY.
FARO Edge ScanArm, FARO, Lake Mary, Fla.
Mimics, version 17.0, Materialise, Plymouth, Mich.
Geomagic Studio, 3D Systems, Rock Hill, SC.
3-matic, version 9, Materialise, Plymouth, Mich.
Python script, Python Software Foundation, Wilmington, Del.
COVTEST, PROC GLIMMIX, SAS, version 10.4, SAS Institute Inc, Cary, NC.
References
1. Hui C, Pi Y, Swami V, et al. A validation study of a novel 3-dimensional MRI modeling technique to identify the anatomic insertions of the anterior cruciate ligament. Orthop J Sports Med 2016;4:2325967116673797.
2. Chang EY, Moses DA, Babb JS, et al. Shoulder impingement: objective 3D shape analysis of acromial morphologic features. Radiology 2006;239:497–505.
3. Prevrhal S, Fox JC, Shepherd JA, et al. Accuracy of CT-based thickness measurement of thin structures: modeling of limited spatial resolution in all three dimensions. Med Phys 2003;30:1–8.
4. Viegas SF, Hillman GR, Elder K, et al. Measurement of carpal bone geometry by computer analysis of three-dimensional CT images. J Hand Surg Am 1993;18:341–349.
5. Lalone EA, Willing RT, Shannon HL, et al. Accuracy assessment of 3D bone reconstructions using CT: an intro comparison. Med Eng Phys 2015;37:729–738.
6. Zhang ZL, Cheng JG, Li G, et al. Detection accuracy of condylar bony defects in Promax 3D cone beam CT images scanned with different protocols. Dentomaxillofac Radiol 2013;42:20120241.
7. Bois AJ, Fening SD, Polster J, et al. Quantifying glenoid bone loss in anterior shoulder instability: reliability and accuracy of 2-dimensional and 3-dimensional computed tomography measurement techniques. Am J Sports Med 2012;40:2569–2577.
8. Fitzwater KL, Marcellin-Little DJ, Harrysson OLA, et al. Evaluation of the effect of computed tomography scan protocols and freeform fabrication methods on bone biomodel accuracy. Am J Vet Res 2011;72:1178–1185.
9. Pinto JM, Arrieta C, Andia ME, et al. Sensitivity analysis of geometric errors in additive manufacturing medical models. Med Eng Phys 2015;37:328–334.
10. Chen X, Zimmerman BK, Lu XL. Determine the equilibrium mechanical properties of articular cartilage from the short-term indentation response. J Biomech 2015;48:176–180.
11. Varga P, Schefzig P, Unger E, et al. Finite element based estimation of contact areas and pressures of the human scaphoid in various functional positions of the hand. J Biomech 2013;46:984–990.
12. Márquez-Florez K, Vergara-Amador E, de Las Casas EB, et al. Theoretical distribution of load in the radius and ulna carpal joint. Comput Biol Med 2015;60:100–106.
13. Majors BJ, Wayne JS. Development and validation of a computational model for investigation of wrist biomechanics. Ann Biomed Eng 2011;39:2807–2815.
14. Klennert BJ, Ellis BJ, Maak TG, et al. The mechanics of focal chondral defects in the hip. J Biomech 2017;52:31–37.
15. Hui-Hui W, Dong W, An-Bang M, et al. Hip joint geometry effects on cartilage contact stresses during a gait cycle. Conf Proc IEEE Eng Med Biol Soc 2016;2016:6038–6041.
16. Katz MA, Beredjiklian PK, Bozentka DJ, et al. Computed tomography scanning of intra-articular distal radius fractures: does it influence treatment? J Hand Surg Am 2001;26:415–421.
17. Leventhal EL, Wolfe SW, Moore DC, et al. Interfragmentary motion in patients with scaphoid nonunion. J Hand Surg Am 2008;33:1108–1115.
18. Márquez-Florez K, Vergara-Amador E, Gavilan-Alfonso M, et al. Load distribution on the radio-carpal joint for carpal arthrodesis. Comput Methods Programs Biomed 2016;127:204–215.
19. Soslowsky LJ, Flatow EL, Bigliani LU, et al. Articular geometry of the glenohumeral joint. Clin Orthop Relat Res 1992;285:181–190.
20. Breit S, Pfeiffer K, Pichler R. Use of a 3D laser scan technique to compare the surface geometry of the medial coronoid process in dogs affected with medial compartment disease with unaffected controls. Vet J 2010;185:285–291.
21. Wilson LAB, Humphrey LT. A virtual geometric morphometric approach to the quantification of long bone bilateral asymmetry and cross-sectional shape. Am J Phys Anthropol 2015;158:541–556.
22. Trinh NH, Lester J, Fleming BC, et al. Accurate measurement of cartilage morphology using a 3D laser scanner. In: Beichel RR, Sonka M, eds. Computer vision approaches to medical image analysis. Berlin: Springer-Verlag, 2006;37–48.
23. DeVries NA, Gassman EE, Kallemeyn NA, et al. Validation of phalanx bone three-dimensional surface segmentation from computed tomography images using laser scanning. Skeletal Radiol 2008;37:35–42.
24. Aubin CÉ, Dansereau J, Parent F, et al. Morphometric evaluations of personalised 3D reconstructions and geometric models of the human spine. Med Biol Eng Comput 1997;35:611–618.
25. Clark JM, Huber JD. The structure of the human subchondral plate. J Bone Joint Surg Br 1990;72:866–873.
26. Dunlap AE, Mathews KG, Walters BL, et al. Three-dimensional assessment of the influence of juvenile pubic symphysiodesis on the pelvic geometry of dogs. Am J Vet Res 2018;79:1217–1225.
27. Harris WH, Rushfeldt PD, Carlson CE, et al. Pressure distribution in the hip and selection of hemiarthroplasty, in Proceedings. 3rd Open Sci Meet Hip Soc, 1975;93–98.
28. Baratz ME, Des Jardins J, Anderson DD, et al. Displaced intra-articular fractures of the distal radius: the effect of fracture displacement on contact stresses in a cadaver model. J Hand Surg Am 1996;21:183–188.
29. Van Dessel J, Nicolielo LF, Huang Y, et al. Quantification of bone quality using different cone beam computed tomography devices: accuracy assessment for edentulous human mandibles. Eur J Oral Implantol 2016;9:411–424.
30. Cao Q, Zbijewski W, Sisniega A, et al. Multiresolution iterative reconstruction in high-resolution extremity cone-beam CT. Phys Med Biol 2016;61:7263–7281.
31. Liptak JM, Dernell WS, Ehrhart N, et al. Cortical allograft and endoprosthesis for limb-sparing surgery in dogs with distal radial osteosarcoma: a prospective clinical comparison of two different limb-sparing techniques. Vet Surg 2006;35:518–533.
32. Herzberg G, Burnier M, Marc A, et al. Primary wrist hemiarthroplasty for irreparable distal radius fracture in the independent elderly. J Wrist Surg 2015;4:156–163.
33. Beckmann J, Stengel D, Tingart M, et al. Navigated cup implantation in hip arthroplasty. Acta Orthop 2009;80:538–544.
34. Ast MP, Nam D, Haas SB. Patient-specific instrumentation for total knee arthroplasty: a review. Orthop Clin North Am 2012;43:e17–e22.
35. Cohnen M, Poll LW, Puettmann C, et al. Effective doses in standard protocols for multi-slice CT scanning. Eur Radiol 2003;13:1148–1153.
36. Brooks RA, Chiro GD. Beam hardening in x-ray reconstructive tomography. Phys Med Biol 1976;21:390–398.
37. Joseph PM, Ruth C. A method for simultaneous correction of spectrum hardening artifacts in CT images containing both bone and iodine. Med Phys 1997;24:1629–1634.
38. Tan Y, Kiekens K, Welkenhuyzen F, et al. Simulation-aided investigation of beam hardening induced errors in CT dimensional metrology. Meas Sci Technol 2014;25;064014.
39. Zhang X, Li L, Zhang F, et al. Improving the accuracy of CT dimensional metrology by a novel beam hardening correction method. Meas Sci Technol 2015;26:1–11.
40. Krumm M, Kasperl S, Franz M. Reducing non-linear artifacts of multi-material objects in industrial 3D computed tomography. NDT Int 2008;41:242–251.
41. Kohonen I, Koivu H, Vahlberg T, et al. Total ankle arthroplasty: optimizing computed tomography imaging protocol. Skeletal Radiol 2013;42:1507–1513.
42. Rathnayaka K, Sahama T, Schuetz MA, et al. Effects of CT image segmentation methods on the accuracy of long bone 3D reconstructions. Med Eng Phys 2011;33:226–233.
43. Eggers G, Klein J, Welzel T, et al. Geometric accuracy of digital volume tomography and conventional computed tomography. Br J Oral Maxillofac Surg 2008;46:639–644.
44. Van den Broeck J, Vereecke E, Wirix-Speetjens R, et al. Segmentation accuracy of long bones. Med Eng Phys 2014;36:949–953.
45. McCann L, Ingham E, Jin Z, et al. An investigation of the effect of conformity of knee hemiarthroplasty designs on contact stress, friction and degeneration of articular cartilage: a tribological study. J Biomech 2009;42:1326–1331.
