A simplified approach to define cervical vertebral levels in spinal cord MRI studies

Abstract

Background and Purpose

Spinal cord (SC) cross‐sectional areas (CSAs) assessed with MRI have proven to be extremely valuable imaging markers in several diseases. Among the challenges is the delineation of vertebral levels to determine level‐dependent changes in cord atrophy. With this study, we aimed to (1) test the hypothesis that there is proportionality in the position of the first six intervertebral discs and the length of the upper portion of the SC and (2) show that a proportionality approach can simplify the CSA assessment across vertebrae offering good reliability.

Methods

Forty‐six volunteers underwent standard T2‐weighted and T1‐weighted cervical SC MRI acquisitions. The distance between the obex and the intervertebral discs (from C2‐C3 to T1‐T2) was measured on the T2‐weighted acquisitions of the entire cohort. In a test‐retest experiment on 12 subjects, the % disc position values were used to define vertebral levels, and a comparison was performed with manual vertebrae assignment in terms of mean CSA and its coefficient of variation.

Results

The mean upper cord length for the cohort was 144.0 ± 13.1 mm. The discs’ level % position in the upper cord was found to be fairly consistent, with standard deviations of 0.8%‐1.7%. The mean vertebral CSA obtained with the proportionality method was substantially equivalent to the manual approach in terms of mean CSA values and test‐retest reliability.

Conclusions

With this study, we propose a proportionality method for the assignment of cervical SC vertebral levels that can simplify the processing of MRI datasets in the context of CSA measurements.

INTRODUCTION

The past few years have witnessed an exponential increase of interest in the assessment of spinal cord (SC) atrophy using MRI, in both the research and the clinical setting. SC cross‐sectional areas (CSAs) and volumes have been proven to be extremely valuable prognostic and predictive (and potentially diagnostic) imaging markers in several diseases affecting the SC, such as multiple sclerosis, motor neuron diseases/amyotrophic lateral sclerosis, traumatic SC injury, and degenerative cervical myelopathy. ¹ , ² , ³ , ⁴

The cervical portion of the SC is by far the region most frequently assessed, since thoracic and lumbar regions pose even more difficult technical challenges. ⁵ When limited acquisition time is available or for technical reasons (eg, coil sensitivity, subject motion), the regions adjacent to either the C2‐C3 intervertebral disc or the C3 vertebral body have become gold standard in the field, not only when reporting SC area/volume measurements but also when reporting other quantitative metrics (eg, those derived from diffusion and magnetization transfer imaging protocols). ⁵ In some situations, however, it can be informative to report the MRI‐derived metrics for all the different cervical SC vertebral levels. For example, motor neuron diseases/amyotrophic lateral sclerosis are very heterogeneous, and the levels of the SC around the cervical enlargement can be affected in different ways in the various phenotypes and disease stages. ⁶ In traumatic SC injury and degenerative cervical myelopathy, it can be very informative to assess not only the level of the lesion/compression, but also regions above and below it. ⁷

Semiautomatic software such as the Spinal Cord Toolbox (SCT) (https://spinalcordtoolbox.com) is today available to help with the vertebral‐level assignment task, ⁸ but the processing can be very time consuming for large datasets, it can fail when image quality is suboptimal, and technical expertise may be required that is not always available. Moreover, a few studies have suggested that, when reporting CSA values, the use of an intrinsic neuroanatomical reference may be more reliable than the use of vertebral body/skeleton regions as reference. ⁹ , ¹⁰

The first aim of this study was to test in a cohort of 46 healthy subjects the hypothesis that there is proportionality in the position of the first six intervertebral discs and the length of the upper portion of the SC (hereafter defined “upper cord” length, measured from obex, a clearly identifiable anatomical landmark, to the T1‐T2 intervertebral disc). Previous studies assessed the location of nerve rootlets and vertebral bodies along the healthy human SC, but did not specifically test this hypothesis. ¹¹ , ¹²

The second aim of the study was to provide a “proof of concept,” testing on 12 healthy subjects in a test‐retest experiment whether CSA assessment using the proportions defined in the first part of the study (hereafter defined “proportionality method”) can provide the same reliability of a more time‐consuming manual vertebrae definition.

METHODS

Subjects

A total of 46 volunteers (20 males, 26 females, age 39.9 ± 14.0) with no history of neurological disorders were enrolled in the study. The Committee on Human Research at our institution approved the study protocols. Written informed consent was obtained from all participants. IP, a neuroradiology resident with more than 5 years of expertise in MRI, inspected all the acquired images to exclude MRI incidental findings and major alterations of spondylosis. Twelve subjects (5 males, 7 females, age 34.4 ± 9.2) out of the 46 underwent MRI between October and November 2017 using a Siemens 3 Tesla (T) Skyra (Siemens Healthineers) and a Philips 3T Ingenia (Royal Philips), while the remaining 34 subjects were scanned during 2023 using only the Skyra scanner.

MRI acquisition

All subjects enrolled in the study underwent a standard cervical sagittal T2‐weighted (T2‐w) 2‐dimensional standard turbo spin echo acquisition (34 subjects on the Siemens scanner—main parameters: in‐plane resolution = 0.72 × 0.72 mm², matrix = 320 × 240, 15 slices 2.4‐mm thick, echo time [TE] = 88 ms, repetition time [TR] = 3000 ms, integrated parallel acquisition techniques [iPAT] = 2, acquisition time = 45 seconds; 12 subjects on the Philips scanner—main parameters: in‐plane resolution = 0.47 × 0.47 mm², matrix = 384 × 384, 17 slices 3‐mm thick, TE = 90 ms, sensitivity encoding acceleration factor 2, acquisition time = 2 minutes 48 seconds). The 12 subjects who underwent the T2‐w acquisitions on the Philips scanner also underwent a sagittal T1‐weighted (T1‐w) 3‐dimensional magnetization‐prepared rapid gradient‐echo acquisition of the SC on the Siemens scanner (voxel size 1 × 1 × 1 mm³, matrix 320 × 260, 192 slices per slab, TR = 2000 ms, inversion time [TI] = 1000 ms, TE = 3.72 ms, flip angle = 9, iPAT = 2, acquisition time = 4 minutes 44 second) repeated twice in a test‐retest experiment (with subject off the scanner bed in‐between sessions).

Image processing/analysis

All the image processing steps and measurements were performed by a neuroradiologist (IP).

On all the T2‐w raw acquisitions from the entire cohort of 46 subjects, the distance between the obex and the intervertebral discs (down to disc T1‐T2) was manually measured along a visually defined centerline that followed the physiological curvature of the SC (example in Figure 1A) using the software Jim (v.9, Xinapse Systems Ltd, West Bergholt, United Kingdom, https://www.xinapse.com). ¹³

FIGURE 1.

Example of measurement performed on the T2‐weighted images (A: raw image, B: straightened image). In Panel A, distances measured along the centerline are reported with different colors (green: obex/C2‐C3; pink: C2‐C3/C3‐C4; yellow: C3‐C4/C4‐C5; blue: C4‐C5/C5‐C6; light blue: C5‐C6/C6‐C7; red: C6‐C7/C7‐T1; white: C7‐T1/T1‐T2), while in Panel B, the green dotted line represents the slice including the obex, the white dotted line represents the slice of the T1‐T2 disc, and the orange dotted lines mark slices correspondent to the other intervertebral discs.

Since software commonly used in the field (SCT and Jim) rely on a cord‐straightening step in advanced processing pipelines that require coregistration to an atlas (eg, SCT's “sct_label_vertebrae”) or between different timepoints (eg, Jim's “cord follow‐up”), we wanted to verify if and how measuring disc positions directly on the straightened cords would affect the results. First, we identified the center of the cord in each subject's sagittal T2‐w image by using the “sct_get_centerline” tool of SCT (v.6.1, Montreal, QC, Canada, https://spinalcordtoolbox.com). Second, we straightened the SC using the “sct_straighten_spinalcord” SCT tool. The distance from the obex to the intervertebral discs (down to disc T1‐T2) was measured again by counting slice numbers on the straightened cords (example in Figure 1B).

Using the “Multi‐Planar Reconstruction” Jim tool, all 24 T1‐w acquisitions (12 subjects × 2 sessions) were cropped from the obex to T1‐T2 and reoriented (maintaining the pixel size) so that the centerline of the SC was perpendicular to the resampling box at the top of the image and along a coronal plane as much as possible. SC CSA of each slice (corrected for the angulation between the SC and each slice orientation) was calculated with Jim over the entire resampled portion, using the “refine center‐line marker positions” option to limit operator‐dependent errors in marker positioning.

On all the 24 T1‐w images, the distance between the obex and the intervertebral discs was measured along the centerline as done for the T2‐w acquisitions. The average CSA per vertebral level (from C3 to C7) was computed by averaging CSA of the slices corresponding to the vertebral body as manually identified (method1) or by using the novel proportionality method (disc position expressed in % relative to the upper cord length obtained from the raw T2‐w images of the entire cohort: method2).

Statistical analysis

All the statistical analyses were performed using JMP Statistics (v17.0.0, SAS Institute, Cary, NC, www.jmp.com) software.

To determine how the straightening process can potentially affect the disc‐level measurement, Mann‐Whitney U‐test and Pearson's correlation were used to compare disc level positions (expressed as % of the upper cord length) determined on the raw T2‐w images and on the straightened T2‐w images for the entire cohort (n = 46). A Bonferroni corrected alpha of .05/6 = .0083 (where 6 is the number of discs assessed) was considered as threshold for significance in these tests.

The average CSA for each of the vertebral levels measured on the T1‐w images on the group of 12 subjects with method1 and method2 was compared using Pearson's correlation coefficients and paired t‐tests on both the test and retest acquisitions. For the latter, a Bonferroni corrected alpha of .05/10 = .005 (5 vertebral levels × 2 acquisitions) was considered as threshold for significance. Bland‐Altman plots for the tests were also produced.

For each of the two vertebral assignment methods, the average coefficient of variation on the group [coefficient of variation (COV) = 100 × (difference/mean)] of CSA was computed for each subject and each vertebral level.

RESULTS

The average upper cord length (distance from obex to T1‐T2 disc) for the 46 subjects ranged from 121.5 to 167.3 mm (mean 144.0 mm, SD 13.1 mm). The distance from the obex to each vertebral disc (from C2‐C3 to T1‐T2) was measured on the T2‐w images on the raw and straightened images. The average position of the discs for the two methods (expressed as % value of upper cord length) is reported in Table 1.

TABLE 1.

Discs’ level % position in the upper cervical cord.

Disc level	Raw data	Straightened data	p (Mann‐Whitney U‐test)	Pearson's r (p‐value)
C2‐C3	32.4% ± 1.7%	32.5% ± 1.6%	.711	.78 (<.001)*
C3‐C4	44.1% ± 1.7%	43.7% ± 1.6%	.329	.82 (<.001)*
C4‐C5	54.9% ± 1.5%	54.5% ± 1.4%	.349	.66 (<.001)*
C5‐C6	65.4% ± 1.4%	65.1% ± 1.5%	.371	.70 (<.001)*
C6‐C7	75.7% ± 1.2%	75.1% ± 1.3%	.049	.72 (<.001)*
C7‐T1	86.7% ± 0.8%	86.3% ± 0.9%	.038	.63 (<.001)*

Note: Position of the six intervertebral cervical discs (form C2‐C3 to C7‐T1) expressed as a percentage of the obex to T1‐T2 disc distance as measured on the raw and straightened T2‐weighted images (46 subjects). All the data represent mean ± standard deviation unless otherwise indicated. A symbol * next to the values underlines significative differences and significative correlations (alpha Bonferroni corrected = .0083).

The disc level % position was found to be fairly consistent on the 46 subjects, with a standard deviation (raw images) ranging for 0.8% for the C7‐T1 disc to 1.7% for the C2‐C3 disc.

The % position of the six intervertebral cervical discs (form C2‐C3 to C7‐T1) did not statistically differ when measured on the raw or straightened T2‐w images (Table 1). The % position of each intervertebral discs measured on the raw and straightened T2‐w images strongly correlated (Pearson's r ranged from .63 for the C7‐T1 disc to .82 for the C3‐C4 disc) (Figure 2).

FIGURE 2.

Comparison of vertebral levels position (% relative to the obex to T1‐T2 length), as measured along the centerline on the raw T2‐weighted (T2‐w) images (y‐axis) and on the straightened T2‐w images (x‐axis) for the entire cohort of 46 subjects.

The average CSA measured on the T1‐w images for the five vertebral bodies defined with method1 (manual definition) or method2 (proportionality method, % reported in the second column of Table 1; for example, for each subject, the C3 vertebral body was defined as the slices included from 32.4% to 44.1% of the obex to T1‐T2 length) is reported for the test and retest sessions in Table 2, together with the Pearson's coefficient comparing the two methods, the p value for the paired t‐tests, and the CSA COV between test and retest sessions.

TABLE 2.

Cross‐sectional area test‐retest experiment with two different vertebral‐level assignment methods.

Vertebral level	CSA test method1	CSA test method2	Pearson's r (test)	Paired t‐test p (test)	CSA retest method1	CSA retest method2	Pearson's r (retest)	Paired t‐test p (retest)	COV method1	COV method2
C3	83.27 ± 5.97	82.75 ± 5.72	.997*	.0035*	83.50 ± 5.37	82.82 ± 5.33	.996*	.0004*	1.17%	0.92%
C4	88.93 ± 5.79	88.42 ± 5.58	.995*	.0116	88.74 ± 6.00	88.26 ± 5.85	.997*	.0079	1.74%	1.72%
C5	89.75 ± 4.55	90.04 ± 4.68	.990*	.1650	88.43 ± 5.32	88.82 ± 5.41	.977*	.2737	2.12%	2.22%
C6	84.65 ± 6.76	84.35 ± 6.85	.960*	.5931	83.51 ± 7.49	83.87 ± 7.31	.939*	.6350	2.33%	2.41%
C7	71.42 ± 8.21	70.99 ± 8.50	.983*	.3584	69.73 ± 8.55	69.68 ± 7.62	.921*	.9594	4.20%	2.94%

Note: For the five vertebral levels the table reports: (1) the cross‐sectional area (CSA) expressed as square millimeters measured with method1 (manual vertebral‐level assignment) and method2 (proportionality method) in the test and retest sessions on the T1‐weighted images of the 12 subjects. All the data represent mean ± standard deviation unless otherwise indicated; (2) the Pearson's r coefficients for the correlations of CSA obtained defining vertebral levels with the two methods; (3) the p values for the paired t‐test comparisons between the two methods; and (4) the average coefficient of variation (COV) between the test and retest session for CSA measured with the two methods. A symbol * next to the values underlines significative correlations and differences (alpha Bonferroni corrected = .005).

The CSA obtained at each vertebral level with method2 and method1 (Table 2) provided very similar mean CSA values for each vertebra, and the correlation of values obtained with the two methods for the 12 subjects was very strong (Pearson's coefficients at all levels were >.921, and always >.977 at the C3, C4, and C5 vertebral levels) (Figure 3). With the paired t‐test comparisons, a small bias in the CSA estimate using the two methods was detected at C3 (Table 2; Figure 4).

FIGURE 3.

Correlations between mean cross‐sectional area (expressed as squared millimeters) obtained for the 12 subjects on the five vertebral bodies with the manual method (method1) and the proportionality method (method2). Left: test session; right: retest session. The 95% interval level is indicated as an ellipsoid.

FIGURE 4.

Bland‐Altman plots for average vertebral cross‐sectional area obtained for the 12 subjects in the test session (retest session plots are similar and not reported). Data are expressed as square millimeters. Continuous red lines represent mean differences, while dashed red lines delimit the upper and lower 95% interval. Method1: manual method; method2: proportionality method.

COVs of measurements obtained with method2 were similar to method1 at the C4, C5, and C6 levels and lower at the top and bottom vertebral body (C3 and C7).

DISCUSSION

With this study, we tested and confirmed the hypothesis of a well‐defined proportionality in the position of the first six intervertebral discs in relation to the length of the upper portion of the SC across different subjects. We have also verified that the straightening process does not substantially affect the proportionality assumption despite potential distortions of the cord. Therefore, we believe that our proposal could be easily implemented in the main available SC software. The finding about disc position proportionality in subjects with different cervical cord length is an interesting result, since it could potentially be used in radiological reports, for example, to define the position of a lesion within the SC and to follow its position in the longitudinal evolution even if some of the discs and vertebral bodies degenerate.

On top of this, in the context of CSA measurements obtained from MRI acquisitions, we showed that obtaining the average CSA per vertebral body using the proportionality method gives very similar mean results if compared with the manual assignment of vertebrae. Not only is the proportionality method simpler than manual or automatic methods (it only requires the identification of the obex—a well‐defined and clearly visible anatomical landmark—and of the T1‐T2 intervertebral disc), but in our test‐retest experiment, the method was even more reliable in terms of COV than the manual vertebrae assignment, therefore making the method particularly appealing for longitudinal studies. The mean COV for the proportionality method was in particular lower for the upper (C3) vertebral body (above the cervical enlargement, where CSA variation with slice is moderate) and the lower (C7) one (where even though the CSA variation with slice is high, ¹⁴ the variability of % disc positioning across subjects was found to be small).

The latter finding is in line with previous work that had suggested using an anatomical landmark instead of disc levels when reporting CSA values for different portions of the SC to improve reliability. ⁹ , ¹⁰ In contrast to previous work, however, we did not measure CSA at fixed distance from an anatomical landmark, but at distances within the upper SC that were rescaled as a function of the cervical cord length.

A small bias in CSA estimates using the two different vertebral‐level assignments was detected at C3 (Table 2; Figure 4). However, on the 12 subjects, this bias is of the same order of magnitude of the test‐retest COV and much smaller than the CSA intersubject variability.

In principle, the proportionality method could also be used to define vertebral levels in acquisitions that only partially include the upper cervical cord (eg, brain images extended to the upper SC); for example, this could be achieved by cropping the images at the obex and at the last intervertebral disc included and accordingly rescaling the proportions reported in Table 1. Moreover, since the simplified approach proposed in this study mainly defines vertebral levels differently, it could be used to report average values per vertebra not only for CSA, but also for any other metrics derived from quantitative cervical MRI acquisitions.

Our study has some limitations that must be acknowledged.

The distance between the obex and the intervertebral discs (from C2‐C3 to T1‐T2) was manually measured on cervical sagittal T2‐w acquisitions with high in‐plane resolution and slice thickness of either 2.4 or 3 mm. Although the obex and the discs were clearly visible in all the images, a partial volume effect could have affected the localization of these structures, especially of the obex; to better localize the latter, the gracile nucleus, which delineates its lateral portions that continue with the central canal at the cervicomedullary junction, was used. The sagittal slice corresponding to the most caudal part of the gracile nucleus was aligned with the SC midplane and included the obex. We believe that this approach has minimized the measurement error.

T2‐w images of the 12 subjects whose T1‐w images were used for the second aim of the study were included in the 46 images used to derive the % values reported in the second column of Table 1. Ideally, it would have been better to use two separate cohorts for the different aims, but we preferred to use as many T2‐w images as possible to provide the reader with numbers derived from a larger cohort.

The number of subjects used to define the position of discs in the proportionality method (46) and the ones scanned with the T1‐w protocol in the test‐retest experiment (12) is not large, and the discs’ level % position we measured could not be appropriate for cohorts with very different sex and age distributions. In particular regarding age, we believe that the findings of this study are valid for an adult population but should be tested in a pediatric population.

Because of the described limitations, an extension of this study on larger and more heterogeneous cohorts is warranted.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Papinutto N, Perretti I, Mallott J, et al. A simplified approach to define cervical vertebral levels in spinal cord MRI studies. J Neuroimaging. 2024;34:639–645. 10.1111/jon.13240