Motion analysis of total cervical disc replacements using computed tomography: Preliminary experience with nine patients and a model

Abstract

Background

Cervical total disc replacement (CTDR) is an alternative to anterior fusion. Therefore, it is desirable to have an accurate in vivo measurement of prosthetic kinematics and assessment of implant stability relative to the adjacent vertebrae.

Purpose

To devise an in vivo CT-based method to analyze the kinematics of cervical total disc replacements (CTDR), specifically of two prosthetic components between two CT scans obtained under different conditions.

Material and Methods

Nine patients with CTDR were scanned in flexion and extension of the cervical spine using a clinical CT scanner with a routine low-dose protocol. The flexion and extension CT volume data were spatially registered, and the prosthetic kinematics of two prosthetic components, an upper and a lower, was calculated and expressed in Euler angles and orthogonal linear translations relative to the upper component. For accuracy analysis, a cervical spine model incorporating the same disc replacement as used in the patients was also scanned and processed in the same manner.

Results

Analysis of both the model and patients showed good repeatability, i.e. within 2 standard deviations of the mean using the 95% limits of agreement with no overlapping confidence intervals. The accuracy analysis showed that the median error was close to zero.

Conclusion

The mobility of the cervical spine after total disc replacement can be effectively measured in vivo using CT. This method requires an appropriate patient positioning and scan parameters to achieve suitable image quality.

Keywords

Adults CT CT high resolution prostheses cervical spine

Cervical total disc replacement (CTDR) is an alternative to anterior fusion (1). The primary goals of total disc replacement are to restore disc height and segmental motion after removing the degenerative disc (2). A secondary goal is to preserve the normal motion of the adjacent levels, which may be theorized to prevent development of adjacent level disease (3). Thus, accurate in vivo measurement of prosthetic kinematics and assessment of implant stability relative to the adjacent vertebrae is critical. The most precise method is radiostereometric analysis (RSA) with an in vitro accuracy of 0.2 degrees for rotation and 0.1 mm for translation (4), with current non-invasive methods being much less accurate. In the lumbar spine, the accuracy of standard lateral radiographs varies from 1.5 to 5 degrees in rotation depending on the different method that is used (5, 6). Yet it is a 2D method that does not measure the 3D motion of the segment. Of the 3D methods that are not invasive, Pearcy et al. (7) described the use of biplanar radiographs in the lumbar spine, and Lim et al. (8) developed a 3D imaging technique using computed tomography (CT) scans in a cadaveric cervical spine. Ochia et al. (9) modified the algorithm and measured in vivo segmental motion of the lumbar spine. Magnetic resonance imaging (MRI) (10) has been used for detecting segmental motion of the spine and the latest is the use of kinematic MRI. All of these non-invasive methods have an accuracy of 1–3 degrees in the spine. However, we have previously reported a new CT-based method, which incorporates CT volume data registration, for detecting segmental movement in the lumbar spine with a rotational accuracy of 0.9 degrees and translational accuracy of 0.6 mm (11, 12).

The aim of this study was to extend this CT-based method for use in the cervical spine by aligning the upper prosthetic component with the CT coordinate system and automating the resulting process. We evaluated the results in terms of the repeatability of measuring rotation and translation in the cervical spine. This was done using both a model and nine patients with CTDR. We also evaluated accuracy of this extended method.

Material and Methods

Nine patients (five men, four women) who had had a primary cervical total disc replacement (Discover artificial cervical disc, Depuy Spine, Inc., Raynham, MA, USA) were included in this study. These patients were the first consecutive nine patients living in the Stockholm area, and were selected from a larger randomized controlled trial comparing the clinical outcome between intervertebral fusion to CTDR median patient age at inclusion was 42 years (range 38–56 years). Three patients had two-level disc replacements. The Karolinska Institute Ethics Committee North approved the study (diary number 2007336313), and informed written consent was obtained from all patients. Three months postoperative, two CT scans of each patient were obtained, one scan with the cervical spine in flexion and one in extension, using a clinical CT scanner (Somatom Definition AS, Siemens, Erlangen, Germany). Table 1 lists the scan details. Each such pair of ‘patient scans’ is termed a ‘patient case’ (Fig. 1a). The patients were placed in the CT scanner on their left shoulder at a 90° angle to the supine position. The head was supported by a soft pillow, and clinically the cervical spine was in neutral position before the scans. The head was then moved voluntarily into maximal flexion and extension respectively, and a CT data volume was acquired in each position. A standard clinical low dose acquisition protocol was used (Table 1). The effective dose was calculated to 0.33 milliSievert (mSv) per scan. Slices were reconstructed at 1.0 mm increments. The x–y pixel size was 0.22 mm for eight patients and 0.29 mm for the other patient. The CT volumes were reconstructed into a matrix of 512 × 512 with 87 slices for eight patients and 57 and 56 slices for the other patient.

Fig. 1
Flowcharts of the various procedures: (a) patient cases; (b) model scans; (c) model pseudo cases; (d) model frame cases

Table 1
Acquisition protocols used for the study

Patients (Siemens) Model (GE)

Slices (n) 64 16

Pitch 0.9 1.375

mAs 47 1.722

Kvp 140 120

Total collimation width (cm) 12 20

For further analysis, a model was constructed, incorporating three human cervical vertebrae and an identical cervical total disc replacement as used in the patients, which allowed orthogonal rotations around three axes (Fig. 2). The model was scanned using a clinical CT scanner (Lightspeed16, GE Healthcare, Waukesha, WI, USA) with a fixed protocol. Slices were reconstructed at 0.5 mm increments. The x–y pixel size was 0.39 mm. The CT volumes were reconstructed into a matrix of 512 × 512 with 302 slices for each scan. The rotation between the upper and lower disc replacement component was initially set to –10° about each axis relative to a neutral position, then the rotation was advanced 5° about each axis (one axis at a time) until a final position of +10° about all three axes was reached. This resulted in 13 unique spatial orientations, termed ‘unique model spatial orientations’ (Fig. 1b). At each unique model spatial orientation, two CT scans were obtained, the second scan was immediately performed after the model was translocated between 10–20° in the scanner, to mimic different patient positions. This resulted in 13 pairs of ‘unique model spatial orientation scans’, where each pair (R1, R2) represents the same unique model spatial orientation R. These scans are then paired, with one representing flexion and the other extension, to produce pseudo patient cases termed ‘model pseudo cases’.

Fig. 2
The cervical spine model allows movement around three orthogonal axes

Procedure

The following is our procedure for measuring the relative kinematics of two cervical spine prosthetic components (an ‘upper component’ and a ‘lower component’) between two CT scans of the same region. Using our image analysis tool, the user (a) visualizes the upper and lower components as 3D isosurfaces (the isosurface level was set at 1250 for the patients and at 2750 for the model studies) in both CT volumes simultaneously, and then (b) manually places landmarks on these isosurfaces at corresponding points in both CT volumes. This completes the manual process. Our image analysis tool automatically performed all remaining steps. It (c) spatially aligns the two CT volumes in 3D such that the upper components coincide, (d) defines a new coordinate system based on the upper components and aligned with the coordinate system of the CT scanner, (e) calculates the orientation of the lower components in each volume in this new coordinate system, and (f) calculates any change in orientation of the lower components (movement) between the two volumes.

Image analysis

All image analysis was performed using a 3D volume fusion (spatial registration) tool, which has been previously described and extensively validated (13–15). This semi-automated tool provides landmark-based fusion of two volumes, registering the ‘target’ volume with the ‘reference’ volume via a variety of 3D transform modules, from simple rigid body, to 3D warping, to user-defined. A graphical interface provides numerous 3D and 2D analysis tools, including those to visualize structures and specify landmarks while viewing from arbitrary positions, with simultaneous display of both reference and target volume information. For our study, the computer's 2D pointing device was used to specify landmarks on isosurfaces, with the software automatically finding the corresponding 3D points. A technical description can be found in previous publications (14–16).

Creating model pseudo cases

The unique model spatial orientation scans were paired up to create ‘model pseudo cases’ with one representing flexion (the fusion reference volume) and the other representing extension (the fusion target volume). As described above, each unique model spatial orientation scan R is represented by two unique model spatial orientation scans (R1, R2), where the only difference is the model position in the scanner. As such, each model pseudo case pairing unique model spatial orientations X and Y, actually consists of two reference volumes, X1 and X2 (both representing a unique model spatial orientation X), and two target volumes, Y1 and Y2 (both representing a unique model spatial orientation Y). Thus, registering either of the two reference volumes with either of the two target volumes should ideally result in identical prosthetic movement. As a result, each model pseudo case (X, Y) can be realized by any of four unique volume pairs, namely (X1, Y1), (X1, Y2), (X2, Y1), and (X2, Y2), which should all have the same results. A pseudo-random number generator was used to create 61 such model pseudo cases. As a case consisted of a two sets of data (a pair) there were 122 sets of data for analysis. For each case the landmarking was repeated, and subsequently the entire procedure, which constituted the repetition study (Fig. 1c).

Landmark selection

To select landmarks on prosthetic components (Procedure steps a and b) for a patient case or a model pseudo case, the user would first visualize the upper and lower components as 3D isosurfaces in both the reference and target CT volumes simultaneously. With the isosurface level set so that voxels less attenuating than metal were not visible (1250 for the patients and 2750 for the model study), a 3D surface identifying the prosthesis itself was obtained (Fig. 3). The upper and lower components were viewed in a standardized way that gave equal access to six elevated pegs on the surface of the prosthesis that face the vertebra. If the patient had a two-level replacement, the most apical prosthesis was chosen. A landmark was manually placed on the tip of each of these six pegs, in a standardized order, for both the upper and lower component. This resulted in an ‘upper landmark set’ and a ‘lower landmark set’ (each with six landmarks), respectively, for each of the two volumes. For each upper and lower landmark set, the fusion tool automatically generated an additional ‘out of plane landmark’ from a vector cross product. This provides a consistent spatial orientation and avoids the possibility of a mirror inversion during subsequent calculations.

Fig. 3
Three-dimensional isosurface display of CTDR in (a) patient and (b) model. Landmark pattern of the upper prosthetic component. Six landmarks designated corresponding features on the back side of the prosthesis. The seventh point is automatically generated from three of the original points to maintain a consistent orientation

To establish repeatability, the same observer in a second trial 24 hours later repeated the above landmark selection procedure for each patient case and each model pseudo case. To provide a more accurate and repeatable reference, the unique model spatial orientation scans were further utilized (now in isolation, rather than paired up as above) by placing a separate set of landmarks on two flat surfaces of the orthogonal frame of the model (Fig. 4). These were grouped as an upper and lower landmark set, mimicking the landmark pattern on the prosthesis, but now on these larger and better defined objects (i.e. planes). This was done for 10 randomly selected scans, producing 10 ‘model frame scans with landmarks’, each representing one CT volume, to then be paired to create model frame cases.

Fig. 4
Three-dimensional isosurface display of the model. Landmarks on the orthogonal frame

Creating model frame cases

The 10 model frame scans with landmarks were paired to create ‘model frame cases’, with one representing flexion (the fusion reference volume) and the other representing extension (the fusion target volume). This differs from the model pseudo cases in that here, the landmark sets were selected on each volume independently, with the volumes and landmark sets later paired to create the cases, whereas in the former, the landmark sets were selected simultaneously on both of the volumes that were already paired as a case. This resulted in 45 model frame cases (all possible unique pairings). In every case, since the frame was unchanged, the upper and lower landmarks should also be unchanged, and each should ideally result in no motion between the scans (Fig. 1d).

Automated processing

Once the upper and lower landmark sets have been created for a pair of CT volumes (Procedure steps a and b), whether for a patient case, a model pseudo case, or a model frame case, the fusion software automatically does the rest (Procedure steps c–f), as follows.

To automatically bring the upper components of the target volume and reference volume into coincidence, a rigid body transform was created (RBT), defined by a singular value decomposition (SVD) derived from the upper landmark set in both volumes. Applying this RBT to the target volume brings it into coincidence with the reference volume at the location of the upper prosthesis component. The new target volume is then saved, creating the ‘registered target volume’. Applying this RBT to the target lower landmark set registers those landmarks into the reference volume coordinate system, allowing direct comparison of this ‘registered target lower landmark set’ with the reference lower landmark set to determine movement. In order to specify movement in a coordinate system aligned with that of the CT scanner and relative to a ‘standard position’ upper component, a ‘standard transform’ was automatically created to achieve the following. Using the upper landmark set from the reference volume, the upper component was placed into a zero rotation (neutral) position. Specifically, an internal orthogonal coordinate base for the upper component was derived from landmarks 1, 5, and 6 (Fig. 3). The internal x-axis was parallel to a vector from 5 to 6, and the z-axis was normal to the plane formed by these three points. Finally, this coordinate system was rotated into alignment with the axes of the CT coordinate system. This standard transform was applied to the reference volume, creating the ‘standard position reference volume’.

Since the upper components from both volumes were registered, the standard transform was applied to the registered target volume, creating the ‘standard position target volume’. The standard transform was then applied to both the reference lower landmark set and registered target lower landmark set, resulting in the ‘standard position reference landmark set’ and the ‘standard position target landmark set’. (Fig. 5) shows the result of the first transformation, followed by the standard position transformation. An RBT derived from the standard position lower landmarks sets was then applied to the ‘standard position target volume’. The transformation matrix was decomposed into (non-unique) Euler angles about each rotation axis and the orthogonal translation (x,y,z) values were extracted. The Euler angles were obtained by decomposing the rotation matrix in the following order: RzRyRx where Rx is the rotation about the x-axis (i.e. the sagittal plane) and was applied first, along the cardinal axes of the vertebrae. This was done for the 61 model pseudo cases, the 45 model frame cases, and in seven of the nine patient cases.

Fig. 5
Registered CT volumes. Patient case. The upper component and the vertebra it is attached to be well aligned, the lower has moved between flexion and extension. This 2D display shows the prosthesis as seen from the side and the front

Loosening evaluation

Finally, the patient case registered volumes were visually reviewed to see if there was any apparent movement between the prosthetic components and adjacent vertebrae or peri-prosthethic osteolysis – as signs of loosening. In the 2D view where all three planes are presented, an overlay function is possible were we can ‘overlay’ the two volumes so that the prosthesis is almost perfect overlaid in the two volumes. We then slowly move from one volume to the overlaid volume and there should be no visual movement between the prosthesis and the vertebra. First we do that on the upper component of the prosthesis and then the lower component of the prosthesis. It has been previously described (17).

Statistics

An ANOVA (18) was done between each patient case and the measurements obtained (i.e. the three angles and the three translation values obtained when each case was processed), and between the paired model pseudo cases in each trial. The data were first analyzed graphically to test for outliers, skewness, unequal variance, and interactions. The first three tests were to show that the data obtained for the measurements in the trials were coherent. The Turkey one degree of freedom test was used to show that the cases and measurements were independent. After the ANOVA was performed, its residuals were accessed graphically to confirm that they were normally distributed. Additionally, the limits of agreements for repeated trials (19) were calculated and displayed graphically. The difference values between all data were tested to determine if the distribution of values were normal. The 95% confidence interval for each limit of agreement was calculated using Student's t-statistic for the appropriate number of degrees of freedom. Repeatability of test results is defined as the precision under conditions where independent tests are conducted with the same method on the same test item in the same laboratory by the same operator using the same equipment within a short interval of time. The repeatability was evaluated using the method outlined in (20, 21). Accuracy is defined as the closeness of agreement between a test result and an accepted reference (a ‘true’) value (21) For the accuracy test using the paired frame model, the measurement data were examined to determine if the data could be considered normally distributed. A box plot was drawn to demonstrate how well all the trials exhibited the expected value of zero. All the statistical calculations were performed using R version 2-12.0.

Results

All model scans could be used, but two patient volumes were unsuitable for analysis because the pegs on the backside of one of the prosthesis, due to partial volume effects, were too poorly visualized in 3D for reliably placing landmarks.

Visual examination of all patient volumes did not show any peri-prosthetic osteolysis or any apparent movement between the prosthetic components and adjacent vertebrae, thus there were no signs of loosening.

The ANOVA showed no significant difference between the cases and the measurements (model cases, trial one [F = 0.10, p = 0.75] and trial two F = 0.12, p = 0.73, patient cases for both trials [F = 1.1, p = 0.36]), but did show some significant differences between the measurements (Euler angles and translation distances as would be expected as all are independent parameters) results – particularly in the patient trials (F = –2.10, p = 0.04) where the distance between the slices was much larger than the x–y pixel size. The ANOVA confirmed that the analysis process was consistent within the cases and across all the cases for the model and patient studies. The plot of the ANOVA residuals showed that they had a normal distribution. The limits of agreements, displayed in Figs. 6–8, showed that almost all the data fell well within the 95% limits. The scales on the graphs have been selected so that the data are displayed consistently for all the angles and distances. The confidence intervals calculated using Student's t-test showed no overlapping between the upper and lower limit of agreement. The repeatability analysis showed 95% of the values were within less than two standard deviations of the mean in accordance with the criterion given in (21) as shown in Table 2. When the measured angle from the CT scans was compared with the rotations in the model the result was consistent and not depending on the size of the angle. This was confirmed by the repeated CT scans and subsequent analysis

Fig. 6
Bland-Altman scatter-plot showing differences between repeated measurements on patient volumes

Fig. 7
Bland-Altman scatter-plot showing differences between repeated measurements on model volumes

Fig. 8
Bland-Altman scatter-plot showing differences when identical repositioning of the prosthesis is measured on different model volumes

Table 2
The repeatability between model and patient

95% limits of agreement on two trials − repeatability

Patients Model

Mean difference 95% upper limit 95% lower limit Repeatability Mean difference 95% upper limit 95% lower limit Repeatability

Coronal (degrees) − 0.08 0.25 − 0.41 0.47 0.04 0.49 − 0.40 0.63

Sagittal (degrees) − 0.36 1.24 − 1.95 2.25 − 0.01 0.76 − 0.78 1.09

Transverse (degrees) − 0.07 1.19 − 1.32 1.78 0.01 0.66 − 0.64 0.92

Coronal (mm) − 0.05 0.23 − 0.33 0.40 0.01 0.21 − 0.19 0.29

Sagittal (mm) 0.06 1.03 − 0.90 1.36 0.00 0.13 − 0.12 0.18

Transverse (mm) 0.06 0.27 − 0.14 0.29 0.01 0.16 − 0.14 0.22

The accuracy with a 95% confidence interval in the model study was for the sagittal plane 0.7° and 0.4 mm for translation, coronal plane 0.4° and 0.2 mm, and for the transverse plane 0.2° and 0.5 mm. The box plots associated with these accuracy tests are given in (Fig. 9). As can be seen from this figure, all the median values lie close to zero. The median movement for the patient was in the sagittal plane for rotation 6.2° and translation 0.1mm, coronal plane 1.6° and 0.6 mm, and for the transverse plane 1.3° and 0.6 mm in translation. All data for patient movement are presented in Table 3.

Fig. 9
Boxplox of accuracy experiment. Note that the sagittal data are plotted with and without scan 5

Table 3
Patients' segmental rotation and translation in C5–C6

Patient Sagittal Coronal Transversal

Rotation (°) Translation (mm) Rotation (°) Translation (mm) Rotation (°) Translation (mm)

1 6.5 − 0.2 − 3.0 − 0.7 − 1.2 0.6

2 8.2 − 0.1 − 1.6 − 2.0 1.1 − 2.1

3 7.7 0.2 − 1.4 0.7 − 0.8 1.2

4 3.5 0 − 1.1 0.3 − 3.2 − 1.0

5 6.2 − 0.4 − 1.7 − 0.7 − 1.4 − 1.1

6 0.7 0.4 − 0.1 0.6 − 1.3 − 0.6

7 1.1 − 1.0 − 2.0 − 0.6 − 2.8 − 0.1

Discussion

In the present study we have shown that this method is able to detect segmental movements in the cervical spine with high accuracy and repeatability in both the model and in patients with total disc replacement. It is not as accurate as RSA in vitro, but in clinical setting the in vivo precision it is almost as good as RSA in the cervical spine (22, 23). However, RSA is an invasive method with it specific problems, the acquisition and analysis is not easy, the method is time-consuming, and the output is purely numeric and can be difficult to interpret. The tantalum markers have to be implanted in the vertebrae and sometimes there is a problem when analyzing the markers that it is too few markers to analyze and/or the spacing of markers is to small witch will create an unstable ridged body. Even though we are using a low dose CT the radiation dose is higher than the RSA dose but the clinical significance of that is unclear according to a AAPM report no 96 (24). Even though this CT-based method is precise it has a potential source of errors. The first, and we believe the largest source of error, is the process of acquiring the CT volumes. The position of the prosthesis in the CT scanner as well as the selected voxel size is important. As the pegs on the prosthesis are very small, partial volume effects (which blur adjacent structures with different electron densities – in this case, the prosthesis and the surrounding bone) will be more pronounced with large voxel sizes. Thus visualization of the pegs will be sharper for smaller voxel sizes than for large voxel sizes. As shown in Fig. 3 it is easier to visualize the tip of the pegs in the model (with a 0.5 mm slice thickness) than in a patient (with a 1 mm slice thickness). However, the x–y resolution is better in the patient studies than in the model studies, as the latter had a larger x–y voxel size (0.39 mm vs. 0.22 mm in most patients). In two cases, it was not possible with the selected CT scanning parameters to make measurements, due to the placement of the patient in the CT scanner resulting in partial volume effects that made it difficult to accurately localize one or more pegs. It can be seen from the results presented in the previous section that the patient studies had a larger rotation error in the sagittal plane as compared to the model studies, but the rotations in the coronal and transverse planes were quite comparable. We interpret this as a direct effect of the voxel's differing in x–y versus z dimensions. Reducing the slice thickness increases the accuracy of localizing the pegs of the prosthesis. In the model the slice thickness of the scans was 0.5 mm, while for patients the slice thickness was 1.0 mm. As a result, there is a correlation between the results of the sagittal plane as compared with the coronal and transverse planes. This problem would be reduced using a CT scanning protocol with smaller slice thickness – without increasing the total radiation dose. We expect advances in CT technology will soon make it possible to reduce the slice thickness even more, thereby increasing this method's accuracy without increasing the total radiation dose. The failure of analyzing two of the patients is due to the problems mentioned above, the slice thickness and the patients positioning in the CT scan. In the beginning of the study the patients were placed on the side with the head resting on a soft pillow. That resulted in an unstable situation with a large rotation of the cervical spine in the coronal plane. When we change to a stiff pillow this problem disappeared. After the changes of slice thickness to 0.6 mm and using stiff pillows we have been able to analyze all 20 prosthesis in an ongoing study.

The second largest cause of error arose during postprocessing of the volumes with our 3D volume fusion tool. There are several methods to identify and place landmarks with high accuracy (25, 26). We have shown in earlier studies that intra- and inter-observer errors are small (16, 17) which is confirmed in this study in the repeatability analysis. Only in scans with the partial volume effect problem discussed above was it problematic to place landmarks. In addition, the process of generating registered volumes induces some loss of information due to the interpolation required to map the data onto the new coordinate matrix. In the lumbar spine studies we lost approximately 1° in accuracy for each registration. During development of the proposed method, we initially tried to place the volumes into a standard position before placing landmarks on them, but a small pilot study using the model data showed that reconstructing the volumes into the standard position matrix lost too much clarity in the images due to the interpolation. Therefore we developed this new process where all landmarks are placed in original volumes before processing them, thus utilizing the full resolution and information content of the original scans.

The results of a test-retest on identical CT volumes in patients and model (Figs. 6 and 7), and thereby is a repeatability study where the test item is the image postprocessing. The range of motion observed in the patients (Table 3) is similar to other studies (1, 27). There are two patients, numbers 6 and 7, with very small movements in all directions. Unfortunately, we did not have any data on patients' pain when these examinations were done. Visual inspection of these volumes showed nothing abnormal and there is no indication in the patient's chart of either of these two patients having more problems with pain or any neurological problems than the other patients in this study.

The test item in Fig. 8 is identical prosthetic movement that is measured in different CT volumes, thus reflecting errors both in CT acquisition and image postprocessing. Most errors stem from the acquisition of the volumes. (Figs. 7 and 8) In both figures the mean error is close to zero, as it should be, but the confidence interval is larger in all directions in Fig. 8.

Accuracy is reflected in Fig. 9. Here the frame of the model, an object much larger than the voxel size was used. The landmarks were further apart and the partial volume effects and the effects of the voxel's dimensions were thereby reduced. As a result errors due to acquisition of the volume were reduced, but not eliminated. The largest error and the only outliers were associated with the sagittal angle and result from scan five, in which part of the frame's isosurface showed partial volume effects due to this face of the frame being nearly parallel to a reconstructed CT slice. It should be noted that in Fig. 9 the sagittal data are plotted with and without scan 5. Also, note that the Euler angles are not unique and that the order of computing the Euler angles was fixed with respect to the CT coordinate system.

With the possibility to visually inspect the volumes in standard positions and from those positions analyze the provocation it was also possible to detect movement not only in patients after CTDR, but also in non-operated patients and patients after fusion. We have done a small pilot study that confirms that this is possible and we are now proceeding to a larger study to answer questions about the state of the fusion. One issue that has to be addressed for these future studies is the definition of an internal coordinate base for the vertebra.

In conclusion, this study has shown a non-invasive CT-based method for detecting 3D segmental movements in the cervical spine after CTDR with both high accuracy and repeatability. The segmental movement can both be analyzed visually and quantified numerically in all three axes. With a low dose scanning protocol the radiation dose is almost as low as for ordinary X-ray. We believe that this method of detecting movement in the spine is useful both in research and for clinical use.

Conflict of interest: None.

	Patients (Siemens)	Model (GE)
Slices (n)	64	16
Pitch	0.9	1.375
mAs	47	1.722
Kvp	140	120
Total collimation width (cm)	12	20

95% limits of agreement on two trials − repeatability
Coronal (degrees)	− 0.08	0.25	− 0.41	0.47	0.04	0.49	− 0.40	0.63
Sagittal (degrees)	− 0.36	1.24	− 1.95	2.25	− 0.01	0.76	− 0.78	1.09
Transverse (degrees)	− 0.07	1.19	− 1.32	1.78	0.01	0.66	− 0.64	0.92
Coronal (mm)	− 0.05	0.23	− 0.33	0.40	0.01	0.21	− 0.19	0.29
Sagittal (mm)	0.06	1.03	− 0.90	1.36	0.00	0.13	− 0.12	0.18
Transverse (mm)	0.06	0.27	− 0.14	0.29	0.01	0.16	− 0.14	0.22

Patient	Sagittal	Coronal	Transversal
1	6.5	− 0.2	− 3.0	− 0.7	− 1.2	0.6
2	8.2	− 0.1	− 1.6	− 2.0	1.1	− 2.1
3	7.7	0.2	− 1.4	0.7	− 0.8	1.2
4	3.5	0	− 1.1	0.3	− 3.2	− 1.0
5	6.2	− 0.4	− 1.7	− 0.7	− 1.4	− 1.1
6	0.7	0.4	− 0.1	0.6	− 1.3	− 0.6
7	1.1	− 1.0	− 2.0	− 0.6	− 2.8	− 0.1

References

Sasso

, Smucker

, Hacker

, Artificial disc versus fusion: a prospective, randomized study with 2-year follow-up on 99 patients. Spine 2007;32:2933–40

Zechmeister

, Winkler

, Mad

. Artificial total disc replacement versus fusion for the cervical spine: a systematic review. Eur Spine J 2011;20:177–84

Fekete

, Porchet

. Overview of disc arthroplasty-past, present and future. Acta Neurochir 2010;152:393–404

Brooks

D SJ

, Ordway

, Evaluation of the accuracy and reliability of radiostereometric analysis in lumbar and cervical spine surgery. 49th Annual meeting of the Orthopaedic Research Society, New Orleans, LA, USA 2003

Panjabi

, Chang

, Dvorak

. An analysis of errors in kinematic parameters associated with in vivo functional radiographs. Spine 1992;17:200–5

Drerup

. Improvements in measuring vertebral rotation from the projections of the pedicles. J Biomech 1985;18:369–78

Pearcy

, Portek

, Shepherd

. Three-dimensional x-ray analysis of normal movement in the lumbar spine. Spine 1984;9:294–7

Lim

, Eck

, An

, A noninvasive, three-dimensional spinal motion analysis method. Spine 1997;22:1996–2000

Ochia

, Inoue

, Renner

, Three-dimensional in vivo measurement of lumbar spine segmental motion. Spine 2006;31:2073–8

10.

Cargill

, Pearcy

, Barry

. Three-dimensional lumbar spine postures measured by magnetic resonance imaging reconstruction. Spine 2007;32:1242–8

11.

Svedmark

, Weidenhielm

, Nemeth

, Model studies on segmental movement in lumbar spine using a semi-automated program for volume fusion. Comput Aided Surg 2008;13:14–22

12.

Svedmark

, Tullberg

, Noz

, Three dimensional movements of the lumbar spine facet joints and segmental movements. In vivo examinations of normal subjects with a new non invasive method. Eur Spine J (in press)

13.

Noz

, Maguire

Jr , Zeleznik

, A versatile functional-anatomic image fusion method for volume data sets. J Med Syst 2001;25:297–307

14.

Olivecrona

, Crafoord

, Olivecrona

, Acetabular component migration in total hip arthroplasty using CT and a semiautomated program for volume merging. Acta Radiol 2002;43:517–27

15.

Gorniak

, Kramer

, Maguire

, Evaluation of a semiautomatic 3D fusion technique applied to molecular imaging and MRI brain/frame volume data sets. J Med Syst 2003;27:141–56

16.

Olivecrona

, Olivecrona

, Weidenhielm

, Model studies on acetabular component migration in total hip arthroplasty using CT and a semiautomated program for volume merging. Acta Radiol 2003;44:419–29

17.

Olivecrona

, Olivecrona

, Weidenhielm

, A new technique for diagnosis of acetabular cup loosening using computed tomography: preliminary experience in 10 patients. Acta Orthop 2008;79:346–53

18.

Faraway

. Linear Models in R. London: Chapman & Hall, CRC, 2004

19.

Bland

, Altman

. Measuring agreement in method comparison studies. Stat Methods Med Res 1999;8:135–60

20.

Bland

, Altman

. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10

21.

Ranstam

, Ryd

, Onsten

. Accurate accuracy assessment: review of basic principles. Acta Orthop Scand 2000;71:106–8

22.

Zoëga

, Kärrholm

, Lind

. One-level cervical spine fusion with or without plate fixation: a randomized study using radiostereometry in 27 cases. Acta Orthop 1998;69:363–8

23.

Zoëga

, Kärrholm

, Lind . Plate fixation adds stability to twolevel anterior fusion in the cervical spine: a randomized study using radiostereometry. Eur Spine J 1998;7:302–7

24.

AAPM report no 96. The Measurement, Reporting, and Management of Radiation Dose in CT-Report of AAPM Task Group 23 of the Diagnostic Imaging Council CT Committee 2008

25.

Pluim

JPW

, Gerritsen

, Likar

, SpringerLink (Online service). Biomedical Image Registration Third International Workshop, WBIR 2006, Utrecht, The Netherlands, July 9–11, 2006, Proceedings. Berlin Heidelberg: Springer-Verlag GmbH, 2006; doi: 10.1007/11784012

26.

Pfluger

, Vollmar

, Wismuller

, Quantitative comparison of automatic and interactive methods for MRI-SPECT image registration of the brain based on 3-dimensional calculation of error. J Nucl Med 2000;41:1823–9

27.

Mummaneni

, Burkus

, Haid

, Clinical and radiographic analysis of cervical disc arthroplasty compared with allograft fusion: a randomized controlled clinical trial. J Neurosurg Spine 2007;6:198–209

95% limits of agreement on two trials − repeatability
	Patients				Model
	Mean difference	95% upper limit	95% lower limit	Repeatability	Mean difference	95% upper limit	95% lower limit	Repeatability
Coronal (degrees)	− 0.08	0.25	− 0.41	0.47	0.04	0.49	− 0.40	0.63
Sagittal (degrees)	− 0.36	1.24	− 1.95	2.25	− 0.01	0.76	− 0.78	1.09
Transverse (degrees)	− 0.07	1.19	− 1.32	1.78	0.01	0.66	− 0.64	0.92
Coronal (mm)	− 0.05	0.23	− 0.33	0.40	0.01	0.21	− 0.19	0.29
Sagittal (mm)	0.06	1.03	− 0.90	1.36	0.00	0.13	− 0.12	0.18
Transverse (mm)	0.06	0.27	− 0.14	0.29	0.01	0.16	− 0.14	0.22

Patient	Sagittal		Coronal		Transversal
	Rotation (°)	Translation (mm)	Rotation (°)	Translation (mm)	Rotation (°)	Translation (mm)
1	6.5	− 0.2	− 3.0	− 0.7	− 1.2	0.6
2	8.2	− 0.1	− 1.6	− 2.0	1.1	− 2.1
3	7.7	0.2	− 1.4	0.7	− 0.8	1.2
4	3.5	0	− 1.1	0.3	− 3.2	− 1.0
5	6.2	− 0.4	− 1.7	− 0.7	− 1.4	− 1.1
6	0.7	0.4	− 0.1	0.6	− 1.3	− 0.6
7	1.1	− 1.0	− 2.0	− 0.6	− 2.8	− 0.1