If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Neurosurgery, Computational Neuroscience Outcomes Center, Brigham and Women's Hospital, Harvard Medical School, Boston, MassachusettsDepartment of Orthopedic Surgery, Hamad General Hospital, Doha, Qatar
Spinal cord delineation is critical to the delivery of stereotactic body radiation therapy (SBRT). Although underestimating the spinal cord can lead to irreversible myelopathy, overestimating the spinal cord may compromise the planning target volume coverage. We compare spinal cord contours based on computed tomography (CT) simulation with a myelogram to spinal cord contours based on fused axial T2 magnetic resonance imaging (MRI).
Methods and Materials
Eight patients with 9 spinal metastases treated with spinal SBRT were contoured by 8 radiation oncologists, neurosurgeons, and physicists, with spinal cord definition based on (1) fused axial T2 MRI and (2) CT-myelogram simulation images, yielding 72 sets of spinal cord contours. The spinal cord volume was contoured at the target vertebral body volume based on both images. The mixed-effect model assessed comparisons of T2 MRI- to myelogram-defined spinal cord in centroid deviations (deviations in the center point of the cord) through the vertebral body target volume, spinal cord volumes, and maximum doses (0.035 cc point) to the spinal cord applying the patient's SBRT treatment plan, in addition to in-between and within-subject variabilities.
Results
The estimate for the fixed effect from the mixed model showed that the mean difference between 72 CT volumes and 72 MRI volumes was 0.06 cc and was not statistically significant (95% confidence interval, –0.034, 0.153; P = .1832). The mixed model showed that the mean dose at 0.035 cc for CT-defined spinal cord contours was 1.24 Gy lower than that of MRI-defined spinal cord contours and was statistically significant (95% confidence interval, –2.292, –0.180; P = .0271). Also, the mixed model indicated no statistical significance for deviations in any of the axes between MRI-defined spinal cord contours and CT-defined spinal cord contours.
Conclusions
CT myelogram may not be required when MRI imaging is feasible, although uncertainty at the cord-to-treatment volume interface may result in overcontouring and hence higher estimated cord dose-maximums with axial T2 MRI-based cord definition.
Introduction
Spinal stereotactic body radiation therapy (SBRT) is increasingly used in the management of metastatic spine tumors.
This strategy aims to increase the probability of local control by delivering large total doses of radiation therapy with steep dose gradients to minimize the damage to surrounding organs at risk, such as the spinal cord.
Hence, it is of paramount importance to adequately delineate the spinal cord to minimize the risk of irreversible myelopathy while maximizing comprehensive tumor coverage to reduce the risk of tumor recurrence.
For spinal cord contouring, magnetic resonance imaging (MRI) sequences such as T2 fast spin-echo, fat-suppression, and constructive interference steady-state have been used to delineate the spinal cord adequately. Nonetheless, MRI has some limitations, which include the need to register it with the planning computed tomography (CT) scan, cerebrospinal fluid (CSF) pulsation artifacts, and geometric distortions.
On the other hand, CT myelography is a good alternative to delineate the spinal cord particularly when MRI is not possible (eg, claustrophobia despite the use of wide-bore/short tunnel scanners, and the presence of certain metal implants such as pacemakers) or in postoperative settings when hardware artifacts impede clear MRI assessment.
However, CT myelogram is a more invasive procedure with attendant risks, including postdural puncture headache, arachnoiditis, and rarely meningitis and is less readily available.
In this study, we evaluated spinal cord contours based on CT simulation with myelogram and compared them to spinal cord contours delineated on fused axial T2 MRI images to determine whether the choice of imaging modality significantly influences spinal cord contouring, including spinal cord volume and contour deviations.
Methods and Materials
After receiving institutional review board approval, we performed a single-center retrospective cohort study of patients who had spinal metastasis treated with spinal SBRT from March 2018 to March 2019 who had CT myelogram with their radiation mapping.
All patients had spinal MRI and CT myelogram fused to CT simulation. Although MRI was performed on multiple scanners, the protocol was similar. Regarding CT myelogram, it was performed with either 10 mL of Omnipaque 180 intrathecal contrast or with 10 mL of Omnipaque 300 intrathecal contrast at the discretion of the performing radiologist. Representative parameters of the T2-weighted sequence and CT are presented in (Table 1). Rigid image registration was performed on the selected anatomic landmarks manually or with assisted automated registration tools using Eclipse (Varian Medical Systems) version 13.6 algorithm. The landmarks were defined on the vertebral bones in the target vertebral body (VB) volume. All fusions were done by a single physicist to reduce registration variations that may affect SC delineation. Each registration was verified by a radiation oncologist.
Table 1Representative parameters of MRI T2-weighted sequence and CT
Variable
Measurement
MRI parameters
TR (ms)
3910-10730
TE (ms)
89-102
Flip angle (degrees)
120-180
FOV (ms)
167-180
Matrix (mm)
256 (except 1 examination got 320 mm)
Slice thickness (mm)
3-5
Slice spacing (mm)
3-6
CT parameters
kVp
120
mAs
Dose modulation
Pitch
0.638-0.75
FOV (mm)
171-225
Matrix (px)
512
Slice thickness (mm)
2-3
Abbreviations: CT = computed tomography; FOV = field of view; kVp = kilovoltage peak; mAs = milliamperes; MRI = magnetic resonance imaging; TE = echo time; TR = repetition time.
All contouring was done on simulation CT scan with a slice thickness of 1 mm. The high-resolution setting was used for all contours. The spinal cord volume was contoured from the superior to the inferior endplate of the target VB volume based on the CT myelogram images and the fused T2-weighted MRI images. Comparing MRI- to myelogram-defined spinal cord, centroid deviations (deviations in the center point of the cord) were assessed at every slice through the VB target volume, and maximum deviations of the cord were assessed; the magnitude of centroid deviation was calculated as:
Upper, central, and lower deviations between MRI-defined and myelogram-defined spinal cord were also calculated for the whole cohort separately in the left/right, anterior/posterior, and superior/inferior axes Fig. 1. Eight providers (5 radiation oncologists, 2 medical physicists, and 1 neurosurgeon) contoured the spinal cords twice, based on CT-myelogram and fused axial T2 MRI.
Figure 1Upper, central, and lower deviations between magnetic resonance imaging-defined and myelogram-defined spinal cord.
To evaluate the 0.035 cc max dose on each individual contour, all 72 contours were rigidly transferred onto their respective patient's treatment plan and calculated using Eclipse's Anisotropic Analytical Algorithm, version 13.6. All treatment plans were optimized and clinically delivered to our patients’ years before this study. The contours used for the optimization were independently done outside of this study and should not favor any of the modalities used. D0.035 cc of the CT Myelogram spinal cord contours and the MR spinal cord contours was obtained to assess the effect of spatial contouring variations on maximum dose variability.
Statistical analysis
The demographics and baseline characteristics of the patients in our sample are shown in Table 2. Because measurements were repeated for each lesion in terms of the methods (CT myelogram and MRI) and the 8 providers, a mixed-effects model was used to account for these multiple sources of variation. Disagreement between the 2 methods could be due to bias or differences in variabilities; CT myelogram and MRI measures were compared by assessing their mean differences (bias), between-subject variabilities, and within-subject variabilities through the mixed-effects model. The bias was assessed through the estimate for the fixed effect for each method. Between-subject and within-subject variabilities were analyzed to see if they were equal between the 2 methods by comparing a model with an unstructured covariance matrix to a model with a compound symmetry covariance matrix through a likelihood ratio test (LRT). The compound symmetry structure imposes all variances to be equal while the unstructured matrix imposes no such constraint and allows each variance to be estimated uniquely from the data. Using the LRT these 2 models could be compared with determine which variance assumption fits the data best, thereby indicating whether variabilities between the 2 methods were equal. Statistical significance was set at α = 0.05. Analyses were conducted using SAS.
Table 2Demographics and clinical/surgical characteristics of 8 patients with 9 spinal metastases receiving spinal SBRT
Age (y)
Sex
Prior spine surgery
Spine hardware at treatment level
Spine site
Epidural tumor extension
Current SBRT dose prescription
Current SBRT in EQD2
Prior radiation therapy to spine site
Radiation therapy dose prescription
Dose in EQD2
72
M
Yes
Yes
T7-T9
No
30 Gy/5 Fx
54 Gy
T7 SRS T8-T10
16 Gy/1 Fx 30 Gy/10 Fx
60 Gy 36 Gy
51
M
Yes
Yes
C5-C7
Yes
30 Gy/5 Fx
54 Gy
None
-
-
75
F
Yes
Yes
T6-T7
No
30 Gy/5 Fx
54 Gy
None
-
-
53
M
No
No
T3-T4
Yes
30 Gy/5 Fx
54 Gy
None
-
-
65
M
Yes
Yes
C5-C7
Yes
30 Gy/5 Fx
54 Gy
C5-C7
20 Gy/5 Fx
28 Gy
44
M
Yes
Yes
T5-T7
Yes
30 Gy/5 Fx
54 Gy
-
-
-
44
M
Yes
Yes
T11-L1
Yes
30 Gy/5 Fx
54 Gy
T11-L1
20 Gy/5 Fx
28 Gy
21
F
Yes
No
T3
Yes
30 Gy/5 Fx
54 Gy
None
30 Gy/5 Fx
54 Gy
62
F
Yes
Yes
T4-T6
Yes
40 Gy/10 Fx
56 Gy
None
-
-
Abbreviations: EQD2 = equivalent dose in 2 Gy fractions; SBRT = stereotactic body radiation therapy; SRS = stereotactic radiosurgery.
During the study period, 8 patients with 9 spinal metastases receiving spinal SBRT were identified and included in our analysis. In our cohort, 3 of 8 patients (37.5%) were female and median age was 57.5 years; 3 of 8 patients (37.5%) had prior spinal radiation therapy, 6 of 8 patients (75%) had spinal hardware, and 6 of 8 patients (75%) had epidural tumor extension as summarized in (Table 2).
Spinal cord volumes for each of these 9 lesions were defined based on (1) axial T2-MRI and (2) CT myelogram by 8 providers, yielding 72 MRI and 72 CT myelogram-based spinal cord volumes. For 5 lesions, CT volumes were nominally greater than MRI volumes; for 4 lesions, MRI volumes were nominally greater than CT volumes (Fig. 2). The estimate for the fixed effect from the mixed model showed that the mean difference between CT and MRI volumes was 0.06cc and was not statistically significant (95% confidence interval [CI], –0.034, 0.153; P = .1832).
Figure 2Mean differences comparing computed tomography to magnetic resonance imaging volumes for each of the 9 lesions contoured by 8 individual providers. Abbreviations: CT = computed tomography; SC = spinal cord.
MRI-defined spinal cord contours had slightly larger upper and lower deviations compared with CT-defined spinal cord contours in the anterior/posterior axis (Table 3). CT-defined spinal cord contours had slightly larger central and upper deviations compared with that of MRI-defined spinal cord contours (central: left/right and superior/inferior axes; upper: left/right and anterior/posterior axes). The mixed model indicated no statistical significance for deviations in any of the axes.
Table 3Intramethod agreement (within-subject) MRI-defined relative to CT-defined spinal cord contours, evaluated at the uppermost, central, and lowest axial slice and reported in (x) the left/right (left defined as positive), (y) anterior/posterior (anterior defined as positive), and (z) superior/inferior (superior defined as positive) axes and dose at 0.035 cc
Variable
CT-myelogram
T2
P value
Volume
0.2704
0.4148
.02
Central
x
0.0008
0.0007
.65
y
0.0019
0.0038
<.01
z
0.0129
0.0456
<.01
Upper
x
0.0013
0.0019
.11
y
0.0036
0.0025
.14
z
-
-
Lower
x
0.0011
0.0009
.29
y
0.0084
0.0112
.21
z
-
-
Dose at 0.035 cc
1.1447
3.1163
<.01
Abbreviations: CT = computed tomography; MRI = magnetic resonance imaging.
Maximum dose estimates at 0.035 cc were available for 8 patients as contoured by 8 providers, for a total of 64 MRI-based spinal cord dose measurements and 64 CT myelogram-based spinal cord dose measurements. The mixed model showed that the mean dose at 0.035 cc for CT-defined spinal cord contours was 1.24 Gy lower than that of MRI-defined spinal cord contours and was statistically significant (95% CI, –2.292, –0.180; P = .0271; Fig. 3).
Figure 3Mean differences (computed tomography minus magnetic resonance imaging, differences in doses in Gy) and 95% confidence intervals computed tomography computed tomography to magnetic resonance imaging maximum spinal cord dose estimates at 0.035 cc for 8 lesions contoured by 8 individual providers. Abbreviation: CT = computed tomography.
The estimated between-subject variabilities for CT and MRI defined volumes were 0.745 and 0.779 (Table 4), respectively. The LRT found no significant difference in the between-subject variabilities of the methods (P = .5769). Estimated within-subject variabilities for CT and MRI were 0.2704 and 0.4149, respectively (Table 3). The LRT was statistically significant indicating that the within-subject variabilities were not equal between the 2 methods with MRI-defined volumes displaying higher variability across raters compared with CT-defined volumes (P = .0160).
Table 4Intermethod agreement (between-subject) MRI-defined relative to CT-defined spinal cord contours, evaluated at the uppermost, central, and lowest axial slice and reported in (x) the left/right (left defined as positive), (y) anterior/posterior (anterior defined as positive), and (z) superior/inferior (superior defined as positive) axes and dose at 0.035 cc
Variable
CT-myelogram
T2
P value
Volume
0.7450
0.7787
.58
Central
x
0.9657
0.9607
1
y
14.5555
14.6942
.11
z
225.83
225.15
.40
Upper
x
1.0304
1.0153
.65
y
16.6186
17.1584
<.01
z
-
-
Lower
x
0.8904
0.8862
1
y
12.1990
12.1645
.75
z
-
-
Dose at 0.035 cc
18.3188
23.2346
.22
Abbreviations: CT = computed tomography; MRI = magnetic resonance imaging.
The between-subject variabilities in the deviations of the 2 methods were not statistically significant for any of the axes except for upper deviation in the anterior/posterior axis (CT, 16.62; MRI, 17.16; P = .0019; Table 4). For within-subject variabilities, there were no statistically significant differences for any of the deviations except for the central deviations in anterior/posterior (CT, 0.0019; MRI, 0.0038; P = .0023) and superior/inferior axes (CT, 0.0129; MRI, 0.0456; P < .0001; Table 3).
The LRT did not show a statistically significant difference in the between-subject variabilities for the dose of the 2 methods (CT, 18.3188; MRI, 23.2346; P = .2207; Table 4). However, there was a statistically significant difference in within-subject variabilities (CT, 1.14; MRI, 3.1163; P < .0001; Table 3), indicating the within-subject variabilities for the dose given at 0.035 cc were not equal between the 2 methods. For most variables in the model, the interobserver variabilities were small.
Given the potential confounding role of surgical hardware in the estimation of lesion volumes, analyses were repeated with spinal hardware also as a covariate in the mixed model. The mean difference between CT volumes and MRI volumes was still 0.06 cc while controlling for spinal hardware (95% CI, –0.040, 0.153; P = .1833). Patients without spinal hardware had 1.1 cc higher volumes for a given method but were not statistically significant (95% CI, –0.848, 3.045; P = .2337).
Discussion
In our comparison of the spinal cord definition based on T2 MRI versus CT myelogram, spinal cord volumes were not significantly different. However, small significant differences were noted in deviations left-to-right (mean deviations ranging from –0.010 to –0.037 cm), and in anterior-posterior deviations (mean deviations ranging from –0.007 to +0.031). These deviations resulted in greater maximum point doses (mean of 1.24 Gy greater) to the spinal cord when it was contoured based on the T2 MRI. This suggests that the aforementioned deviations result in the T2 MRI overestimating the spinal cord at the spinal cord-treatment volume interface (assuming the CT myelogram to be a gold standard for spinal cord at that interface). This finding may be the result of the poorer definition of the spinal cord by the T2 MRI resulting in uncertainty at this interface, in comparison to CT myelogram where the myelogram dye permits greater confidence as to the location of this interface. These findings suggest that CT myelogram, as part of CT simulation, is not necessary when cord visualization is feasible on MRI during SBRT treatment planning and administration and when adequate dose coverage of the target volume can be achieved even with some overestimation of spinal cord (with its resultant limitations to target volume coverage). However, if spinal cord overestimation may result in significant under the coverage of the target volume, then consideration should be given for the addition of CT myelography for spinal cord definition. This is likely to be more of an issue in spinal cord retreatment cases where the spinal cord dose maximums are limited because of prior RT or where there is the proximity of the spinal cord to the target volumes (eg, significant epidural disease). Reassuringly, the T2 MRI does not appear to underestimate spinal cord dose compared with the CT myelogram-based spinal cord. Furthermore, our study includes 7 of 9 cases with hardware in the spine treatment region, and comparisons of spinal cord parameters using MRI T2 sequences on 1.5T machines with hardware suppression protocols versus CT myelogram-based spinal cord definition did not yield findings different from those of the larger cohort. This suggests MRI with hardware suppression protocols can also be used for patients with spine hardware, where MRI imaging is feasible to define the spinal cord.
Imaging is the cornerstone of treatment planning during the selection of patients for spinal SBRT to determine local tumor extent, overall disease burden in the spine, and identify possible SBRT contraindications.
One of the major initial barriers to the adoption of SBRT for spinal indications is the risk of radiation myelopathy, given the high doses and steep dose gradients directly adjacent to the spinal cord.
Chronic progressive radiation myelopathy is one of the most feared complications that generally develops within 9 to 15 months after radiation therapy, with a shorter latent period in patients receiving spine reirradiation and pediatric spine cases.
This has particular implications of myelopathy risk for radiation treatments where the dose per fraction is high and the number of treatments is a few to single fraction regimens.
Unlike traditional radiation therapy, spinal SBRT uses a combination of image guided technologies that permit delivery of treatment plans with steep dose gradients for greater dose delivery for better local control with the sharp fall-off of the dose gradient to minimize damage to adjacent organs at risk.
This technique is critical in treating spinal tumors given local failure can be catastrophic and given the sensitivity of neighboring normal tissues such as the spinal cord where high dose per fraction treatment can cause irreversible myelopathy.
Based on patterns of treatment failure after SBRT, marginal failures immediately at or just beyond the conformal targeted volume have been identified as a primary pattern of recurrence. In particular, epidural disease progression, the site immediately proximate to the spinal cord, is the most common site of failure in the de novo, retreatment, and postoperative settings, highlighting the critical nature of accurate target volume and spinal cord delineation.
proposed that CT images were suboptimal for visualizing the spinal cord because it only permits visualization of the bony spinal canal and not the spinal cord structure contained within the thecal sac. Hence CT-based normal structure delineation employs the larger, bone-defined spinal canal as a proxy for the spinal cord. Given this issue, Geets et al
Inter-observer variability in the delineation of pharyngo-laryngeal tumor, parotid glands and cervical spinal cord: Comparison between CT-scan and MRI.
postulated that CT might overestimate spinal cord dose (ie, the spinal canal) resulting in undercoverage of the targeted tumor, leading to a higher risk of tumor recurrence. To overcome this dilemma, spinal SBRT requires the use of imaging modality that visualizes the cord itself, which is necessary to optimize target volume dose coverage and to protect neural structures from radiation-related injury. Therefore, the mainstays of cord imaging are MRI-based imaging, typically using T2-weighted images, or CT myelography.
As a standard imaging modality, MRI has been widely used to delineate neural structures, particularly T2-weighted imaging, which has excellent contrast between the dark spinal cord and the bright CSF. Some shortcomings of MRI include the need to register it with the planning CT scan, CSF pulsation artifacts, and geometric distortions particularly in the presence of metal-related artifacts.
Possible steps may be taken to overcome metal artifacts in MRI studies include optimal patient positioning, determining imaging plane and section thickness and field of view, using metal artifact reduction sequences or short tau inversion recovery sequences, increasing bandwidth and echo train length, and using small voxel volumes. However, many of these steps result in long scan times that are not clinically practical or feasible.
Although CT myelography is an invasive procedure and infrequently used because of potential complications such as spinal headache, radiculopathy, cerebral edema, or spinal subarachnoid hematoma,
it may be regarded as a suitable alternative to MRI in selected cases. The administered contrast fills the CSF space around the spinal cord, resulting in good contrast between the spinal cord and CSF so that the spinal cord contours can be delineated. Sudha et al
assessed the role of CT myelography in sparing the spinal cord during definitive radiation therapy in vertebral hemangioma. They found that CT myelogram was useful in cases where the tumor margin is in close relation to the spinal cord and in cases of spinal implants that result in MRI distortions. Similarly, Thariat et al
reported better physical resolution with CT myelography compared with 3-dimensional Fast Imaging Employing Steady-state Acquisition (FIESTA) MRI performed in 11 patients. According to the authors, a more accurate spinal cord delineation was yielded using CT myelography, particularly in the presence of metal artifacts. Our findings similarly suggest that for cases in which the proximity of the spinal cord to the target volume would prohibit adequate coverage of nearby target volumes, a CT myelogram should be considered. However, if the spinal cord is adequately visualized with MRI, and the proximity of the spinal cord to the treatment volume is sufficiently large to allow for adequate dose coverage of the target volume while ensuring the spinal cord dose maximum adheres to established dose constraints, then MRI alone can be considered. Of course, patients unable to have an MRI because of contraindications (eg, MRI incompatible implants) require a CT myelogram for spinal cord definition.
Reasons for the difference in spinal cord contours between CT myelography and T2-weighted MRI may be related to characteristics of the imaging techniques, which are important to consider troubleshooting how to improve accuracy in defining the spinal cord. First, differences in spinal cord contouring may be related to volume-averaging in the craniocaudal direction on the MRI scans, which have rectangular voxels that are much larger in slice thickness than the voxels on the CT myelogram. The larger slice thickness results in more volume averaging of spinal cord and CSF, thereby somewhat blurring the spinal cord and CSF margin. A potential imaging method for overcoming this issue is the use of 3-dimensional T2-weighted sequences with small isotropic voxels, such as T2 SPACE.
Our group is currently pursuing a study to compare standard T2-weighted MRI versus T2 Space MRI sequences for spinal cord definition to determine whether this modality may aid in reducing uncertainty at the spinal cord-CSF interface. Differences may also be related to parameters of the CT myelogram, for example, the smoothing algorithm used in the reconstruction based on either a soft tissue versus a bone reconstruction kernel. Using a bone reconstruction kernel for the CT myelogram should lead to sharper borders in differential tissues, such as at the spinal cord-CSF (containing myelogram dye) interface. Notably, we compared our spinal cord volumes based on the bone reconstruction kernel images versus the soft tissue reconstruction kernel images CT myelogram and we did not find any significant differences in spinal cord volumes, suggesting that this imaging factor is not a dominant cause of spinal cord volume error.
Limitations
Our study is limited by its retrospective design. It is also limited by single-institution experience and a small sample size. However, data from multiple experienced neuroradiologists, radiation oncologists, and neurosurgeons allowed for clinically relevant analysis of the 2 radiologic modalities compared here. Future work should assess planning parameters for each modality, including target volume dose coverage and normal tissue dose metrics. Future work should also assess alternative approaches to MRI-based spinal cord imaging such as 3D T2-weighted imaging (T2 SPACE), examining its effect in comparison to standard 2D T2-weighted images, CT myelography, and on ultimate target volume and normal tissue dose parameters.
Conclusions
Our data comparing T2 MRI-based and CT-myelogram-based spinal cord definition suggest that although spinal cord volumes are not different between these modalities, there are small, with no meaningful differences in the volumes or deviations. We saw some differences in the variabilities in some of the axes, these minor deviations result in a slight overestimation of spinal cord maximum doses by an average of 1.24 Gy when the spinal cord is based on T2 MRI imaging. These data suggest that spinal cord contouring based on T2 MRI is adequate for SBRT planning where there is sufficient space between the spinal cord and target volumes to allow for adequate target coverage while adhering to spinal cord dose constraints. However, if there is the risk for target volume under coverage (eg, due to prior RT or close proximity of the spinal cord to the treatment volume), then a CT myelogram should be considered. Future studies are needed to determine the optimal modality that minimizes imaging distortions and optimizes true cord definition.
Inter-observer variability in the delineation of pharyngo-laryngeal tumor, parotid glands and cervical spinal cord: Comparison between CT-scan and MRI.
Sources of support: This work had no specific funding.
Disclosures: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
All data generated and analyzed during this study are included in this published article and its supplementary information files.
00264-0&id=gr1.jpg){kind=link}
00264-0&id=gr2.jpg){kind=link}
00264-0&id=gr3.jpg){kind=link}