Abstract
Background
Multiplanar reconstructions of computed tomography (CT) scans can alleviate issues with bone or joint positioning during scan acquisition. The repeatability of these reconstructions is dependent on human operators applying reconstruction criteria, and therefore is subject to error, which could affect measurement reliability for angular or spatial measurements made for orthopaedic surgery.
We describe a method for quantifying interreconstruction variability numerically and graphically using metadata from the CT header to find vectors describing reconstruction axis alignment. The approach is demonstrated using 3 sets of computed tomographic reconstructions of 24 vulpine femorotibial joints.
Results
Vectors describing axis alignments permitted identification and subsequent analysis of deviations from optimal alignment between reconstruction sets. For the worked example, alignment deviations equivalent to femoral abduction/adduction were nearly twice those for extension/flexion, and simulation of the effects of these deviations on measurements closely matched published data.
Conclusions
The method presented here is straightforward and permits numerical and graphical analysis of reconstruction variability. Reconstruction alignment variability should be considered before adopting new reconstruction criteria for clinical use, and evaluated whenever there is suspicion that reconstruction variability could unduly influence subsequent measurements. These evaluations may help drive improvements in reconstruction criteria. The methods described here could also be employed for comparing patient positioning between scans and between different scan modalities.
Background
Ideally, computed tomographic (CT) scans of a patient’s bones and joints should be obtained with the region of interest optimally aligned with the gantry. Optimal alignment can be difficult to achieve in distinct human medical populations [1,2,3]. Similar issues are faced in the veterinary field, where CT is becoming increasingly popular for evaluation of orthopaedic disease, because of anatomic constraints on limb positioning. Veterinary applications include measurement of femoral anteversion (torsion) [4], femoral varus [4, 5], tibial torsion [4], tibial tuberosity alignment [4, 6], and tibial tuberositytrochlear groove (TTTG) offset distance [7]. Diagnostic measurements in veterinary orthopaedics frequently necessitate multiplanar reformatting of the native scans because adequate alignment of the patient’s bones or joints with the gantry can be difficult or impossible to achieve.
Limited data on reconstruction variability are available. In humans, positional variability does not have a consistent effect, with tibial torsion measurements reportedly resilient [8] and TTTG measurements quite sensitive [9]. The poor reported agreement between CT and MRI scanning for TTTG [10, 11] and between repeated CT measurements of TTTG [12] could in part be due to positional variability between scans. These issues are likely similar for veterinary uses of CT. A single observer experimental investigation into interreconstruction repeatability of TTTG measurements made in the red fox (Vulpes vulpes) found a marked increase (1.6 mm vs 0.5–0.7 mm) in the repeatability coefficient for measurements made in the xy plane between reconstructions as opposed to within reconstructions [7]. The repeatability coefficient is an estimate below which the maximum absolute difference between two paired measurements should lie in 95% of instances [13]. If reconstruction variability were negligible, repeatability coefficient values would be expected to be similar in all cases: the observed discrepancy suggests that reconstructions were not identical and that these differences impacted subsequent measurements.
In order to reduce measurement variation between reconstructions or scans at different time points, better reconstruction or positioning criteria could be used. However, this presupposes a technique to evaluate reconstruction or interscan variability and thus compare the effects of different criteria on the repeatability of multiplanar reconstructions or patient positioning. Multiplanar reconstructions are made by rotations of the native scan about the x, y and z axes. Information about the final orientation of the reconstruction relative to the original axes is stored in the DICOM (Digital Imaging and Communications in Medicine) header for the reconstructed images as a series of directional cosines which describe the orientation of the first row and first column of each reconstruction slice [14].
Although reconstruction variability appears initially to be a threedimensional problem, errors relating to rotation about one axis (the viewing axis) can usually be disregarded because measurements are typically made in planes normal to this axis, either on a single slice or by projection of points from one slice onto another. For any set of reconstruction criteria and a particular measurement, the issue of reconstruction variability may therefore be reduced to consideration of rotational errors about the other two axes, which together define the measurement plane.
A method has been described using the separation angle to quantify reconstruction variability, but this has the disadvantage of combining errors in rotation about two axes into a single angular description [7]. This leads to an inevitable loss of information and hampers efforts to identify the underlying causes of reconstruction variability and thus ameliorate them.
This paper presents an improved method to quantify reconstruction variability in multiplanar reformatted CT scans, and is intended as an aid to researchers in investigating reconstruction variability and developing improved reconstruction criteria.
Results
For the three reconstructions of the single femorotibial joint shown in the worked example section, the directional cosines of interest for the zaxis, along with the equivalent angular deviations from the zaxis in the directions of the x and y axes are shown in Table 1. A stereographic plot of the three points is shown in Fig. 1.
When this process was continued for all 24 femorotibial joints, variances \( {\sigma}_{z_x}^2 \) and \( {\sigma}_{z_y}^2 \) were 1.35 deg^{2} and 0.49 deg^{2} respectively. Covariance was − 0.07 deg^{2}, yielding an ellipse angle of − 4.4° (using Eq. 5). After scaling the output of Eq. 4 for a 95% error ellipse (k = 2.448), the semimajor (along the x axis) and semiminor (along the y axis) axes were 2.8° and 1.7°, respectively. The 95% error ellipse area was 15.3 deg^{2}. An orthographic plot of these data and the associated ellipse is shown in Fig. 2.
The full set of direction cosines and associated data is available, demonstrating application of steps 5–9 and 10–14 described above [15].
Extrapolation to TTTG study
The mean TTTG measurement was 0.23 mm medial to the orientation of the trochlear sulcus, with a range of 1.9 mm, corresponding to an error due to reconstruction variability of ±0.95 mm. This range was essentially unaltered by zeroing the tilt semiaxis (simulating no rotation about the xaxis): in contrast, zeroing the spin semiaxis to zero (simulating no rotation about the yaxis) reduced this range to 0.1 mm.
The magnitude of the z_{x} angle, the major contributor to interreconstruction variability, was positively and significantly correlated with the interreconstruction variance in TTTG measurements obtained previously [7], Spearman’s rho 0.53 (p < 0.001).
Discussion
The method for quantifying interreconstruction variability demonstrated here is straightforward, especially with the aid of a spreadsheet, and has the advantage over the only previously described method [7] of not reducing two variables to a single variable. The ellipse parameters and graphical analysis enable ready appreciation of differences in axis alignment between reconstructions made from the same native scans. The method described here could readily be adapted for evaluation of positional variability between native scans obtained at multiple time points, or positional variability between scan modalities, both of which may influence subsequent measurement repeatability [10,11,12].
We are not aware of previous reports focusing on interreconstruction variability. Possible reasons include belief that interreconstruction variability or its effects are insignificant, or a lack of appreciation that interreconstruction variability can occur or can be quantified. Although some indication of the effect of reconstruction variability can be obtained by comparing measurements from different reconstructions, if these are performed by the same observer there may be learning bias in the obtained measurements, and if performed by different observers it is impossible to distinguish between the effects of reconstruction and observer. It is important to be able to determine at which level measurement differences arise (i.e. during multiplanar reconstruction or during reading), since this will alter the approaches needed to minimize these differences. If differences are mostly due to reconstruction variability, optimizing reconstruction criteria and focusing training here is more relevant than focusing on landmark identification for the actual measurements, and vice versa.
We recommend a combined graphical and numerical evaluation of angular error data, in order to exclude extreme outliers that may result in ellipse parameter bias and to fully appreciate the data. Graphical analysis allows rapid appreciation of areas of concern, for example marked deviations along one axis or covariation of the data. Numerical analysis provides data which may be used to compare with future results obtained using improved reconstruction criteria or different methods. The choice of error ellipse size is left to the individual researcher, although the 95% error boundary is a generally accepted limit. The approach used here assumes random variability and normal distribution. The accuracy of the estimates for the ellipse axes (equivalent to the standard deviation) is therefore dependent on the chisquared distribution with a degree of freedom equal to nm1, where n is the number of individuals or samples and m is the number of reconstructions per individual or sample [16]. Using 3 reconstructions per limb as shown here reduced the range of the 95% confidence interval for the ellipse axes by a factor of 0.8 compared to using only 2 reconstructions: using 4 would further reduce the range by a factor of 0.9. Individual researchers will need to determine their accuracy requirements in relation to the extra workload involved in creating additional reconstructions and the number of individuals or samples available.
The data presented here are orthographically projected. In an orthographic projection, points on the spherical surface are projected onto the associated plane along lines normal to this plane. With increasing angular deviation of these points from the axis connecting the pole in contact with the plane to the antipodean pole, the orthographic projection underestimates the distances from the contact pole to the points. In contrast, a stereoscopic projection, in which points are projected along a line from the antipodean pole passing through the points and onto the projection plane, results in slight overestimation. However, at the angles at which researchers are likely to be investigating reconstruction variability, these errors are likely to be insignificant.
In the example shown, the data variance along the x axis was almost three times that along the y axis, indicating some difficulties in precisely controlling rotation about the y axis during multiplanar reconstruction of the native scans. This rotation can be considered equivalent to abduction and adduction of the femur.
Graphical analysis showed most reconstructions were within 3°4° of the intended neutral alignment along the xaxis. This is not dissimilar to reported deviations from intended alignment with the CT gantry for human femora and knee joints of between 1°11° of adduction [2, 9]. Ultimately, the significance of this variability will be decided by the clinician or surgeon in light of its effect on subsequent measurements. Reconstructions in the red fox TTTG study were performed to strict criteria and double checked using maximum intensity projection images [7], but these precautions were inadequate to completely constrain either positional or measurement variability.
Additional safeguards on reconstruction variability, a change in reconstruction criteria or a change in measurement technique would appear necessary to reduce measurement variability. A major problem with this series of reconstructions was the use of landmarks solely confined to the trochlea to define the sagittal plane, in contrast with typical medical practice. This was a deliberate choice to test a hypothesis that the tibial tuberosity is neutrally aligned with the trochlea in the canid femorotibial joint (i.e. that the local TTTG value is zero). In contrast, the coronal plane was defined by landmarks at either end of the femoral diaphysis, which markedly constrained tilt variation. Whilst use of similarly placed landmarks for defining the sagittal plane would have likely constrained yaxis rotation similarly, values for TTTG measurement would then become dependent on distal femoral varus or valgus. This would complicate interpretation and require subsequent correction for femoral morphology, and was the reason for restricting the landmarks so severely for these reconstructions.
Simulated rotation and projection of coordinates retrieved from the TTTG study CT scans confirmed the findings of the graphical analysis, and provided additional information. Reconstruction variability alone could produce TTTG measurement errors of up to ±0.95 mm, based on a 95% error ellipse, and this was mostly attributable to spin errors, or rotation about the yaxis. In the actual TTTG study [7], the interreconstruction repeatability coefficient (representing the maximum expected difference between 95% of paired observations) was ±1.6 mm, which fits very well with the predictions from the simulation, given intrareconstruction repeatability coefficients of ±0.5–0.7 mm (as reconstruction variability of 0.95 + intrareconstruction variability of 0.5–0.7 = interreconstruction variability of 1.45–1.65). Here we assume the intrareconstruction values to represent solely observer error, and the interreconstruction values to represent the sum of observer error and reconstruction variability. The magnitude of the spin error (z_{x}) was positively correlated with interreconstruction variability, supporting the importance of controlling reconstruction variability in order to minimise measurement error. In this specific instance, use of the approach outlined here in combination with the simulation would likely have resulted in further efforts to reduce interreconstruction variability prior to starting the measurement process.
The relevance of similar error magnitudes will vary with the type of measurement, landmarks involved, and patient population. In one study in humans, a mean femoral adduction of 6.6° resulted in mean femoral anteversion measurements over 4° greater than with neutral alignment [2]. Clinically, patient measurements are often compared to predetermined cutoff values in order to determine the need for surgical intervention, and the degree of correction required. If similar variability occurred in CT scans of dogs with medial patellar luxation, for example, it is clear that misclassification as either requiring or not requiring femoral detorsion could occur, along with the potential for over or undercorrection of abnormal anteversion. The researcher may approach this problem in one of two ways. Where there is preexisting literature regarding the effect of alignment on measurement error, likely thresholds for reconstruction error may be directly inferred and used to compare with the error ellipse obtained above. Where such information does not exist, a trial and error approach may be employed using the described macro [17], to find input limits at which the simulated measurement error becomes unacceptably large. These input limits may then be used in place of literaturederived values.
Methods
Principals
The method outlined below is relatively straightforward and can be accomplished with the aid of a spreadsheet. A workbook containing spreadsheets to perform steps 5–9 described below is freely available [18].
Extracting directional cosines from the DICOM header
The initial steps require calculation of the rotations performed during multiplanar reconstruction and can be summarised as:

1.
Acquire native scans of the region of interest from multiple individuals.

2.
Perform multiplanar reconstruction according to a set of defined criteria.

3.
Repeat step 2 at least twice so that for each individual there are at least 3 versions of the multiplanar reconstruction for analysis.

4.
Group these reconstructions in sets, within which each individual features only once.

5.
For each reconstruction in each set, extract the DICOM header data from the Image Orientation (0020, 0037) tag. This may be achieved using the DICOM header viewer available in most viewing software or in bulk using a macro in ImageJ [19], available in the aforementioned workbook. The Image Orientation tag records the direction cosines for the first row and first column of each image slice in the stack, corresponding to unit vectors along the x, y axes.

6.
Calculate the direction cosines for the z axis (see Table 2). This can be accomplished directly using in the aforementioned workbook, which both converts the text output of the macro into individual direction cosines and also calculates the z axis components.
For an investigation of positional variability between native scans at different time points, steps 1–3 may be achieved by acquisition of these scans for multiple patients.
Defining the rotational matrix and its inverse
Each set of components for each axis (e.g. x_{x}, x_{y}, x_{z}) represents a unit vector describing the orientation of that axis. Since the three axes are orthogonal, their vector components, taken together, represent an orthogonal rotation matrix R:
Matrix R describes the transformation from the initial patient orientation to the reconstruction orientation. The transpose of this matrix, R^{− 1}, enables back rotation from the reconstruction axes to the original orientations of the native scan. Because initial patient orientation will differ between native scans, this source of variation needs to be removed. This can be achieved by finding mean vectors for each individual across reconstructions, deriving R^{− 1} from these mean vectors, and then rotating the separate vectors for each reconstruction. These steps can be performed automatically in the spreadsheet PrecisionCalculator within the PrecisionAnalyser workbook.

7.
The mean vectors for each individual across their reconstructions are found by averaging each set of axis components, e.g.

8.
The back rotation matrix R^{− 1} for each individual is found by arranging the mean vector components in rows to form a 3 × 3 matrix, such that

9.
Multiply each set of vectors for each reconstruction by R^{− 1} to centre the x, y, z vectors on their respective primary axes, removing variation due to initial positioning.
Identifying the vectors of interest
The primary axis of interest (viewing axis) must now be identified. The distribution of vectors around the viewing axis represents the misalignments of this axis introduced during multiplanar reconstruction. For measurements made in the sagittal, coronal and transverse planes, the primary axes of interest will typically be the x, y, and z axes respectively. Of the three components defining this primary axis of interest, the two representing deviations along the orthogonal axes should be retrieved. For example, if the measurement of concern is obtained from transverse slices, the primary or viewing axis will be z, and the two orthogonal axes for which data should be retrieved will be z_{x} and z_{y}. The distribution of vectors along z_{x} represents the effects of rotation of the viewing axis about the yaxis, whereas the distribution along z_{y} represents rotation of the viewing axis about the xaxis, as will be seen later.
Graphical and numerical evaluation
These components can be converted into angular deviations along the two orthogonal axes by finding the arccosine of each component of interest before plotting using an orthographic projection. As shown in Table 3, at restricted angular deviations likely to be experienced in practice there is little difference between true arc length along the surface of the unit sphere and the orthographic projection. It should be noted that deviations along one axis represent rotational error about the perpendicular axis.
The data can now be analysed using Model II regression techniques [20] to find ellipse axes and related characteristics describing the distribution of the points.

10.
Find the variances (\( {\sigma}_a^2,{\sigma}_b^2 \)) for the components of interest (e.g. x_{y}, x_{z}) across all reconstructions, and the Pearson correlation coefficient ρ between these components.

11.
Calculate the covariance as σ_{ab} = ρ ∙ σ_{a} ∙ σ_{b} and derive the eigenvalues of the variancecovariance matrix \( \left(\begin{array}{cc}{\sigma}_a^2& {\sigma}_{ab}\\ {}{\sigma}_{ab}& {\sigma}_b^2\end{array}\right) \) using the formula

12.
The major and minor semiaxes of the error ellipse corresponding to the errors about the two reconstruction axes of interest are found as the square roots of λ_{1} and λ_{2} derived using Eq. 4.

13.
The ellipse area A can be calculated as A = π ∙ r_{a} ∙ r_{b}.

14.
If there is substantial covariance, the ellipse angle θ (representing anticlockwise rotation from the horizontal semimajor axis) can be calculated as
These semiaxis values define a 39% error ellipse: a scaling factor k can be used to encompass other error probabilities [21] using \( k=\sqrt{2\ln \left(1p\right)} \) where 0 < p < 1. These parameters enable plotting of the error ellipse, comparison between different reconstruction criteria, and identification of potentially problematic reconstruction axes.
Worked example
Three sets of multiplanar reformatted CT scans produced for a published study of TTTG measurement in 24 red foxes (Vulpes vulpes) [7] were used as the source material [22]. Cadavers were obtained from a commercial unit following euthanasia by electrocution in accordance with Danish law and were unaffected by orthopaedic disease of the hind limbs based on physical inspection and radiography. Institutional ethical approval was obtained. During scanning of the hind limb, anatomical constraints resulted in obliquity of both femora and tibia relative to the gantry. Three reconstructions were produced for each scanned limb by a single operator according to previously defined criteria [7]. Briefly, one plane passed through the centres of the femoral head and medial condyle and parallel to the caudal aspects of both femoral condyles; the second plane was aligned orthogonally and along the femoral trochlear sulcus; the third plane was aligned coorthogonally to the first two. Based on the published result showing a 3fold change in measurement repeatability across reconstructions compared to within reconstructions, there was a concern that reconstruction variability could have played a significant role in measurement variation.
Extracted directional cosines, the calculated z axis cosines and the mean vectors (using Eq. 2) for one individual and three reconstructions are shown in Table 4. These yield the transposed rotation matrix (Eq. 3) for this individual of:
This matrix is then applied to the individual axis vectors: as an example, the reverse rotation of the zaxis vector from the first reconstruction for this individual is shown.
Extrapolation to TTTG study
To estimate the effect of reconstruction variability on TTTG measurement accuracy in the initial study [7], a further analysis was performed using a custom Visual Basic for Applications macro in Excel: a workbook containing this macro and guidelines for its use is freely available [17].
Using freely available software (ImageJ, [19]), three dimensional coordinate data representing the locations of the caudal aspects of the lateral and medial femoral condyles, the base of the trochlea sulcus and the tibial tuberosity were transferred to a spreadsheet, along with error ellipse semiaxes (scaled using k = 2.45 for a 95% error ellipse) and angle. Running the macro rotated the coordinate data in 200 steps along the ellipse perimeter, projecting the coordinates onto a reference plane and calculating the TTTG measurement at each step. The TTTG measurement was calculated as the shortest distance between two lines, both perpendicular to a line passing through the femoral condyle projections, passing through either the trochlear sulcus or tibial tuberosity projection. Output data included mean, standard deviation, minimum, maximum and range of TTTG measurement for each reconstruction. Runs were repeated with each of the semiaxes alternately set to zero, to estimate the individual contributions of errors in these directions.
Conclusion
In conclusion, we present a technique that permits quantification of reconstruction variability, the results of which seem to match measurement variation seen in practice. We suggest that reconstruction variability should be evaluated before adopting reconstruction criteria for clinical use, whenever there is suspicion that reconstruction variability could unduly influence subsequent measurements. These evaluations may help drive improvements in reconstruction criteria, permit comparisons between different reconstruction criteria or methods, and can form the basis of geometric modelling of measurement errors due to reconstruction variability. The methods described here could also be employed for comparing patient positioning between scans and between different scan modalities.
Availability of data and materials
The datasets generated and analysed during the current study are available in the figshare repository:
PrecisionAnalyser worksheet, https://doi.org/10.6084/m9.figshare.1544294, Microsoft Excel worksheet, GNU General Public Licence [18].
PrecisionAnalyser with worked example, https://doi.org/10.6084/m9.figshare.1544295, Microsoft Excel worksheet, GNU General Public Licence [15].
CT reconstructions and extracted header metadata, https://doi.org/10.6084/m9.figshare.3435410, CCBY licence [22].
CTreconSim – a simulator of reconstruction induced measurement variability,
https://doi.org/10.6084/m9.figshare.1337839.v6, Microsoft Excel worksheet, GNU General Public Licence [17].
Abbreviations
 CT:

Computed tomography
 DICOM:

Digital Imaging and Communications in Medicine
 MRI:

Magnetic resonance imaging
 TTTG:

Tibial tuberositytrochlear groove distance
References
Davids JR, Marshall AD, Blocker ER, Frick SL, Blackhurst DW, Skewes E. Femoral anteversion in children with cerebral palsy  assessment with two and threedimensional computed tomography scans. J Bone Jt Surg  Am Vol. 2003;85A:481–8.
Olesen TH, Torfing T, Overgaard S. MPR realignment increases accuracy when measuring femoral neck anteversion angle. Skelet Radiol. 2013;42:1119–25.
Riccio AI, Carney J, Hammel N, Stanley M, Cassidy J, Davids JR. Threedimensional computed tomography for determination of femoral anteversion in a cerebral palsy model. J Pediatr Orthop. 2015;35:167–71.
Barnes DM, Anderson AA, Frost C, Barnes J. Repeatability and reproducibility of measurements of femoral and tibial alignment using computed tomography multiplanar reconstructions. Vet Surg. 2014;44:85–93.
Oxley B, Gemmill TJ, Pink J, Clarke S, Parry A, Baines S, et al. Precision of a novel computed tomographic method for quantification of femoral varus in dogs and an assessment of the effect of femoral malpositioning. Vet Surg. 2013;42:751–8.
Towle HA, Griffon DJ, Thomas MW, Siegel AM, Dunning D, Johnson A. Pre and postoperative radiographic and computed tomographic evaluation of dogs with medial patellar luxation. Vet Surg. 2005;34:265–72.
Miles JE, Jensen BR, Kirpensteijn J, Svalastoga EL, Eriksen T. Measurement repeatability of tibial tuberositytrochlear groove offset distance in red fox (Vulpes vulpes) cadavers. Am J Vet Res. 2013;74:888–94.
Eckhoff DG, Johnson KK. Threedimensional computed tomography reconstruction of tibial torsion. Clin Orthop. 1994;302:42–6.
Yao L, Gai N, Boutin RD. Axial scan orientation and the tibial tubercletrochlear groove distance: error analysis and correction. Am J Roentgenol. 2014;202:1291–6.
Camp CL, Stuart MJ, Krych AJ, Levy BA, Bond JR, Collins MS, et al. CT and MRI measurements of tibial tubercletrochlear groove distances are not equivalent in patients with patellar instability. Am J Sports Med. 2013;41:1835–40.
Schoettle PB, Zanetti M, Seifert B, Pfirrmann CWA, Fucentese SF, Romero J. The tibial tuberositytrochlear groove distance; a comparative study between CT and MRI scanning. Knee. 2006;13:26–31.
Lustig S, Servien E, Selmi TAS, Neyret P. Factors affecting reliability of TTTG measurements before and after medialization: a CTscan study. Rev Chir Orthop Reparatrice Appar Mot. 2006;92:429–36.
Bland JM, Altman DG. Statistics notes: measurement error and correlation coefficients. Br Med J. 1996;313:41–2.
NEMA. Image Plane Module. DICOM PS3.3 2016e  Information Object Definitions. 2016;:438–40. http://dicom.nema.org/medical/dicom/current/output/pdf/part03.pdf. Accessed 6 Dec 2016.
Miles JE. PrecisionAnalyser Worked Example; 2015. https://doi.org/10.6084/m9.figshare.1544295.
Sokal RR, Rohlf FJ. Confidence limits for variances. In: Sokal RR, Rohlf FJ, editors. Biometry. 4th ed. Basingstoke: W.H. Freeman and Company; 2012. p. 167–8.
Miles JE. CTreconSim; 2016. https://doi.org/10.6084/m9.figshare.1337839.v6.
Miles JE. PrecisionAnalyser; 2015. https://doi.org/10.6084/m9.figshare.1544294.
Schneider CA, Rasband WS, Eliceiri KW. NIH image to ImageJ: 25 years of image analysis. Nat Methods. 2012;9:671–5.
Sokal RR, Rohlf FJ. Model II Regression. In: Sokal RR, Rohlf FJ, editors. Biometry. 4th ed. Basingstoke: W.H. Freeman and Company; 2012. p. 535–44.
Harter HL. Circular Error Probabilities. J Am Stat Assoc. 1960;55:723–31.
Miles JE. Fox femorotibial joint reconstructions; 2016. https://doi.org/10.6084/m9.figshare.3435410.
Acknowledgements
The authors would like to thank Professors Thomas Eriksen and Fintan McEvoy for their critical input.
Funding
Institutional.
Author information
Authors and Affiliations
Contributions
JM performed data analysis and drafted the manuscript. Both authors were involved in the conceptual design of the study. Both authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
Institutional approval for use of cadaveric animal material was obtained (Ethical administrative committee, Department for Veterinary Clinical Sciences, Faculty of Health and Medical Sciences, University of Copenhagen).
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
About this article
Cite this article
Miles, J.E., Buelund, L.E. Quantifying the variability between multiple multiplanar reconstructions of computed tomography scans. BMC biomed eng 3, 2 (2021). https://doi.org/10.1186/s42490021000477
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s42490021000477
Keywords
 Computed tomography
 Reconstruction
 Skeletal
 Alignment
 Variability