Systematic assessment of inflammation by magnetic resonance imaging in the posterior elements of the spine in ankylosing spondylitis




Magnetic resonance imaging (MRI) is sensitive for scoring inflammatory lesions in the spine, but attention has primarily focused on vertebral bodies, and no study has systematically examined the posterior elements. We aimed to systematically determine the frequency and distribution of inflammatory changes in the posterior elements of the spine using MRI, and to assess the reliability of their detection and their impact on discrimination of spinal MRI.


We scanned 32 patients recruited to placebo-controlled trials of anti–tumor necrosis factor therapy. Inflammatory lesions were detected by systematic review of consecutive sagittal STIR slices of the entire spine. Two readers evaluated pretreatment and posttreatment scans, blinded to treatment and time point. Inflammation was scored dichotomously (present/absent) in each posterior structure. Reproducibility was assessed by calculating random model variance components and generalizability coefficients, and discrimination by using Guyatt's effect size.


Most patients (87.5%) had ≥1 lesion in the posterior elements (mean ± SD number of affected spinal levels per patient 6.7 ± 5.3), and they were detected most frequently in the thoracic spine. Interobserver reproducibility for total lesion count was very good to excellent for lesions in the thoracic spine and transverse and spinous processes. The addition of a simple dichotomous method for scoring posterior element inflammation substantially enhanced the discrimination observed using established MRI methods for scoring vertebral body inflammation.


Inflammatory lesions in the posterior elements were present in the majority of patients with AS, and standard MRI protocols of the spine should be modified to ensure adequate visualization of posterolateral structures.


Magnetic resonance imaging (MRI) is the most sensitive imaging modality available for the detection of inflammatory lesions of the spine and sacroiliac joints. This has been made possible by the advent of T2-weighted MR sequences, such as STIR images, that suppress the signal from bone marrow fat. This allows the visualization of edema, which is seen as a bright T2 signal on a dark background (1). MRI scoring methods, such as the Spondyloarthritis Research Consortium of Canada (SPARCC) spinal inflammation score, have been developed and validated for the assessment of inflammation in the spine (2, 3). These approaches only evaluate lesions in the vertebral bodies, ignoring lesions in the posterior structures of the spine. However, assessment of posterior structures such as facet joints may be of greater clinical significance because these inflammatory lesions may lead to serious functional impairment and disability, and one study has shown that fusion of the facet joints precedes the development of vertebral ankylosis (4).

Plain radiography shows that structural lesions are frequently evident in the posterior elements of the spine, with fusion of the facet joints being the most typical. However, radiographic assessment is difficult due to the complex bony anatomy, and reliable evaluation in the thoracic spine is impossible due to overlapping structures that obscure the posterior elements. Pathologic studies have shown inflammation within facet joints and at entheses related to spinous and transverse processes. Also, MRI studies have described abnormal T2 signals within vertebral pedicles, facet joints, transverse processes, spinous processes, and soft tissues where ligaments attach to bone. Their frequency and distribution in the spine, particularly in relation to inflammatory lesions in the vertebral bodies, have yet to be systematically examined to our knowledge. This may have diagnostic implications for spinal assessment by MRI.

This report describes the first systematic evaluation of the frequency and distribution of inflammatory lesions in the posterior structures of the spine by MRI. We also conducted preliminary assessment of reliability and discrimination, focusing on the detection of lesions on a dichotomous basis. Our aim was to develop a systematic approach to the assessment of inflammation in posterior structures that will facilitate diagnostic ascertainment, and to determine whether the discrimination of the SPARCC instrument for scoring spinal inflammation is enhanced when inflammation in posterior elements is additionally quantified.



We studied 32 patients diagnosed with AS according to the modified New York criteria (5) who were recruited into 3 randomized, phase III, placebo-controlled trials of anti–tumor necrosis factor α (anti-TNFα) therapy in which MRI examination was conducted at baseline and at either 12 or 24 weeks (6–8). Among these patients, 23 (71%) were male, and the mean ± SD age was of 42.3 ± 11.7 years (range 21–70 years) and mean ± SD disease duration was 11.8 ± 11.4 years (range 2–42 years). The mean Bath Ankylosing Spondylitis Disease Activity Index score (9) was 6.1 (range 3.7–8.8) and the mean Bath Ankylosing Spondylitis Functional Index score (10) was 5.8 (range 2.2–8.4). We also included 3 controls with a history of nonspecific back pain. All controls were imaged with an MRI protocol identical to the patients with AS. The study was approved by the University of Alberta ethics committee and all patients provided informed consent.


All MRI of the spine were performed with 1.5T systems (Siemens, Erlangen, Germany) using appropriate surface coils. Sagittal sequences were obtained with 3– 4 mm slice thickness and 11–15 slices acquired. Sequence parameters were 1) T1-weighted spin-echo (repetition time [TR] 517–618 msec, time to echo [TE] 13 msec), and 2) STIR (TR 3000–3170 msec, time to inversion 140 msec, TE 38–61 msec). The field of view was 380–400 mm and the matrix was 512 × 256 pixels. The spine was imaged in 2 parts: the upper half, comprising the entire cervical and most of the thoracic spine, and the lower half, comprising the lower portion of the thoracic spine and entire lumbar spine. The specific MRI parameters for acquiring spine images are provided on our website (

MRI reading exercises.

A unique MRI study number was allocated to each patient and control, thereby ensuring blinding to all patient demographics. Allocation was done by a technologist unconnected with the study using computer-generated random numbers. Assessment was performed on a 3-monitor review station by 2 readers using computer software that is optimal for this type of review (Merge efilm, Wisconsin, MW), and viewing conditions were standardized. Each subject was only identified by the MRI study number, and scans were read in random order by 2 fellowship-trained musculoskeletal radiologists (SMC, SSD). Readers were unaware of the treatments, the time points, and the number of control subjects included in the review. Definitions of abnormalities in the posterior elements were finalized after training, which consisted of an initial pilot study of 15 scans of patients with AS read by the 2 observers, with consensus being attained using a third musculoskeletal radiologist (RGWL). The pilot results were then discarded, and these scans were not used in the main study.

Increased bone marrow signal, denoting inflammation in vertebral bodies and posterior elements, was assessed on STIR sequences. Positive T2 signal for inflammation was defined as a signal that was greater than the signal from the center of an adjacent normal vertebral body. MR images for each patient were evaluated at 25 spinal levels from C1 to S1. We recorded inflammatory signal (dichotomous yes/no) on STIR images at each spinal level using an electronic, online scoring module designed for this study (Figure 1). In the thoracic and lumbar vertebrae, the data collection schematic was designed to separately identify inflammatory change in the vertebral body and 4 components of the posterior structures: 1) the pedicles, 2) the facet joints, 3) the transverse and spinous processes, and 4) the posterior soft tissues (Figure 1B). In the cervical spine, it was not possible to apply these definitions because the individual components of the posterior elements could not be reliably separated. Therefore, the presence of abnormal T2 signal in any posterior element of the cervical spine was recorded dichotomously, treating the posterior arch of the vertebra as a single structure (Figure 1A). Attribution rules were constructed to avoid double counting and to assist with correlation analysis: 1) bilateral lesions were counted as a single positive for that structure, 2) transverse and spinous process lesions at the same level were counted as a single positive for process, 3) lesions in facet joints were assigned to the vertebral level of the inferior facet of the joint (e.g., a lesion in either facet of the T6/7 facet joint was designated as “T7 facet joint positive”), and 4) intertransverse or interspinous soft tissue edema was assigned to the vertebral level above.

Figure 1.

Schematic diagrams from an electronic, online scoring module designed for scoring inflammation in spinal structures. For A, the cervical spine, the observer can record inflammation as present in the vertebral body (vb), posterior elements (pe), and/or posterior soft tissues (st). For B, the thoracic and lumbar spine, the observer can record inflammation as present in the vb, pedicle (ped), facet joints (fac), transverse/spinous processes (proc), and/or posterior st.

The methodology for quantifying vertebral body inflammation according to the SPARCC method has been described previously and includes the evaluation of the 6 most severely affected discovertebral spinal units (6-DVU score), as is recommended in clinical trials, as well as all 23 spinal DVU (23-DVU score), as is recommended for observational studies (3). SPARCC 6-DVU and 23-DVU scores were determined independently and prior to any assessment of posterior element inflammatory lesions.


Descriptive data.

The distribution of inflammatory lesions in the posterior elements according to patient spinal segment, and vertebral level was analyzed descriptively. The data provided represent the mean scores for the 2 readers. We compared the frequency of inflammatory changes in the posterior elements in those with and without inflammation in the adjacent vertebral bodies using Fisher's exact test, because the pilot study showed that vertebral body inflammation and inflammation in the posterior structures was often present simultaneously. When edema was present in the pedicles, it was often unclear whether the primary focus of inflammation was in the vertebral body or the posterior structures. It was therefore important to determine to what degree the finding of inflammatory changes in the different posterior elements occurred, independently of changes in the vertebral bodies.

Reliability data

Analysis of interobserver reliability for the detection of inflammatory changes in the posterior elements addressed several different comparisons in both the baseline and posttreatment MRI, with reliability being addressed according to both total and segmental lesion score per patient and according to detection of lesions (yes/no) at individual spinal levels. We compared 1) the reliability of detection of all lesions in the posterior elements with those in the vertebral bodies; 2) the reliability according to spinal segment; 3) the reliability of the detection of lesions in the 4 different structures comprising the posterior elements (pedicles, facet joints, processes, and soft tissues); and 4) the reliability of the detection of change in the total number of lesions in the vertebral bodies and the posterior elements.

To estimate reliability, we used the computer program GENOVA (Springer-Verlag, New York, NY), which calculates random model variance components within analysis of variance and calculates an overall reliability coefficient, the generalizability coefficient (G-coefficient), that ranges from 0–1 (where 0 = completely unreliable and 1 = maximum reliability). For the first set of analyses of interobserver reliability of total lesion scores, we used a 1-facet random model with a patient × observer (P × O) design with 2 trained observers. We examined the reliability of the pretreatment and posttreatment status scores as well as the reliability of the change scores. For the second set of analyses for each spinal structure, we used a mixed design: patient × observer × spinal level (P × O × L). This design was used to assess the degree of variability attributable to observers when averaged across the spinal levels for each spinal structure. A G-coefficient of ≥0.8 was designated as representing a reasonable estimate of reliability.


We used Guyatt's effect size to assess the discrimination between anti-TNFα therapy and placebo of the SPARCC 6-DVU and 23-DVU scores when inflammation in posterior elements was additionally quantified dichotomously (present/absent) using 2 approaches: an all–posterior elements score (a simple summation of inflammatory lesions in each individual posterior element across the entire spine), or an any–posterior elements score (the score resulting from summation across all 23 spinal levels where inflammation was recorded as being present in any 1 of the posterior element at a particular spinal level). Guyatt's effect size was calculated by dividing the mean change in the anti-TNFα group by the SD of the change in the placebo group. An effect size >0.80 is considered large.


Descriptively analyzed data.

The majority (87.5%) of patients had an inflammatory lesion recorded in at least 1 posterior element, and this was the same proportion that had an inflammatory lesion in at least 1 vertebral body (87.5%). Two patients were noted to have lesions only in the posterior elements, but this was noted in different patients by the 2 readers. The thoracic spine was most frequently affected by inflammation in the posterior elements, with 84.4% of patients having ≥1 such lesion followed by the lumbar spine (46.9% of patients) and then the cervical spine (31.3% of patients). Frequency of inflammation in specific posterior element structures was as follows: 84.4%, 68.8%, 62.5%, and 31.3% of patients had inflammation in ≥1 transverse or spinous process, facet joint, pedicle, or soft tissue, respectively. The mean number of spinal levels with inflammation in the posterior elements per patient was 6.7, this being observed most frequently in the thoracic spine and the transverse and/or spinous processes, and was somewhat less than the mean number of vertebral bodies with an inflammatory lesion, which was 8.4 (Table 1). After treatment with anti-TNFα therapy, inflammation was significantly (P < 0.001) and comparably reduced in both vertebral bodies and posterior elements. Inflammatory lesions in the posterior elements were significantly more likely to be detected in those spinal levels that also had lesions in the vertebral bodies (mean of 4.1 affected levels per patient) as compared with those levels that had lesions only in the posterior elements (mean of 2.6 affected levels per patient) (P < 0.0001). Some subtle lesions were identified in 2 of the control patients by both readers. The majority of these small lesions were in vertebral bodies in which all of them were associated with evidence of disc degeneration and/or small tears of the annulus fibrosus.

Table 1. Distribution of posterior element inflammatory lesions in 32 patients with AS by spinal segment and according to involvement of specific posterior element structures*
 Vertebral bodiesPosterior elementsPedicleFacet jointsSpinous processSoft tissue
  • *

    Values are the mean ± SD number of spinal levels with inflammatory lesions. AS = ankylosing spondylitis; ND = not done.

Cervical spine1.2 ± 2.00.8 ± 1.3NDNDND0.4 ± 0.9
Thoracic spine5.2 ± 4.34.8 ± 3.72.1 ± 2.62.4 ± 2.83.2 ± 2.90.3 ± 1.0
Lumbar spine2.2 ± 2.11.2 ± 1.60.5 ± 1.01.0 ± 1.40.3 ± 0.60.2 ± 0.4
Total8.4 ± 6.76.7 ± 5.32.6 ± 3.33.5 ± 3.83.7 ± 2.90.9 ± 1.5

A total of 800 spinal levels were included in this analysis, of which 252 (31.5%) had inflammatory lesions in the vertebral bodies and 186 (23.2%) had lesions in the posterior elements prior to treatment with anti-TNF agents. The latter were more common in thoracic vertebrae (32.8%) and in transverse and/or spinous processes (18.8%). Affected posterior elements were more frequently detected in the lower half of the thoracic spine (Figure 2). Significant correlations were noted between inflammation in vertebral bodies and in either the pedicles (r2 = 0.55, P = 0.001) or the facet joints (r2 = 0.57, P = 0.0006) at the corresponding spinal level. No significant correlation was noted with inflammation in the transverse and/or spinous processes.

Figure 2.

Three parasagittal STIR images from an MRI scan of the cervicothoracic spine of a 43-year-old man with a long history of back pain and ankylosing spondylitis (TR 3,170 msec, TI 140 msec, TE 38 msec, 4 mm thick). A, intense increased signal intensity (bone marrow edema) in the right transverse process of T10 (short arrow). Other transverse processes are less uniformly affected, such as at T6 (long arrow), where the edema lies eccentrically posteriorly within the process and appears unrelated to the costotransverse joint. B, bone marrow edema extensively affects the thoracic spine but is more prominent in the posterior elements. Facet edema is intense at the T7/8 level (small arrow). The T8 pedicle is edematous related to the zygapophyseal disease, and the T8/9 discovertebral joint appears normal (long arrow). Facet arthropathy at T4/5 (arrowhead) is associated with bone edema, facet erosion, synovial thickening, and periarticular inflammation. The T5 pedicle is edematous; however, definite evidence of discovertebral disease is also present at this level, with bone edema seen in the adjacent T4 and T5 vertebral endplates (curved arrow). C, bone edema can be seen in multiple spinous processes and is often eccentric, as seen inferiorly at T5 (long arrow) and toward the tip at T6 (short arrow). Vertebral bodies are minimally affected by bone marrow edema.

Interobserver reliability of the detection of inflammatory lesions.

The reliability of both status and change scores for the total count of posterior element lesions per patient was very good and comparable with that observed for the vertebral bodies, especially for lesions in the thoracic spine (intraclass correlation coefficient [ICC] 0.91 for change score) and in the pedicles and processes (ICC 0.74 and 0.77 for change scores, respectively) (Table 2). The reliability of the status score was good for the cervical and lumbar spine but was poor for change scores (ICC 0.31 and 0.44, respectively).

Table 2. Interobserver reliability of the total number of inflammatory lesions per patient with AS (n = 32), at baseline and after exposure to anti-TNFα therapy, in the vertebral bodies and PE of the spine*
RegionStatus ICC pretreatmentStatus ICC posttreatmentChange ICC
  • *

    AS = ankylosing spondylitis; anti-TNFα = anti–tumor necrosis factor α; ICC = intraclass correlation coefficient; PE = posterior elements.

Vertebral body0.850.920.88
Total PE0.820.870.85
Cervical PE0.680.470.31
Thoracic PE0.810.900.91
Lumbar PE0.740.910.44
Facet joints0.760.820.69
Soft tissues0.280.810.55

The sources of variation and percentage of total variance averaged across the spinal levels for each spinal structure at baseline are shown in Table 3. The percentage of variance due to the difference in patients was the largest proportion of variance in all structural elements except soft tissue, and variance due to the observer effect was very small (0.0–5.8%). The G-coefficients met the criteria for reasonable reliability (≥0.80) for the vertebral body, posterior element, thoracic posterior element, and pedicle, but not for soft tissue and the cervical posterior element. This can be explained partly by the small number of inflammatory lesions identified for these latter 2 spinal structures, which reduces the variability between patients (Table 1).

Table 3. Components of variance (percentage of total variance) in the detection of inflammatory lesions in the vertebral bodies and posterior elements by 2 observers in 32 patients with AS*
Sources of varianceVertebral bodyTotal PECervical PEThoracic PELumbar PEPedicleFacet jointsSpinal processSoft tissue
  • *

    Values are the percentage of total variance. Sources of variance are averaged across the spinal levels for each structural element. AS = ankylosing spondylitis; PE = posterior element.

  • P:R residual = the variability in the patient–observer interaction confounded by other sources of systematic variability not accounted for or specified in the model and random error variability.


Observer discrimination.

Discrimination between anti-TNFα therapy and placebo was substantially enhanced when inflammation in posterior elements was scored and added to either the 6-DVU or 23-DVU SPARCC scores (Figure 3). This was noted irrespective of whether inflammation was scored in each individual posterior element (all–posterior elements score) or when inflammation was recorded as being present in any one of the posterior elements at the level of an individual spinal unit (any–posterior elements score). Guyatt's effect size was >3 for all scores that included posterior inflammation, indicating very high discrimination between treatment groups.

Figure 3.

Enhancement of the discrimination between anti–tumor necrosis factor therapy and placebo by the addition of a simple dichotomous method for scoring posterior element (PE) inflammation to the Spondyloarthritis Research Consortium of Canada magnetic resonance imaging 6-DVU or 23-DVU. Guyatt's effect size >0.80 represents a large treatment effect.


This systematic evaluation of inflammatory lesions in both anterior and posterior structures of the spine describes several novel observations. First, the majority of patients with AS who were considered candidates for treatment with anti-TNF agents had lesions in the posterior structures of the spine. Second, lesions in the posterior structures occurred almost as frequently as those in the anterior spine, and they were particularly frequent in the thoracic vertebrae and in the transverse and spinous processes. Third, lesions in the pedicles and facet joints occurred in association with lesions in the vertebral bodies, but lesions in the spinous and transverse process were not obviously associated. Fourth, the presence of lesions in the posterior structures could be detected reliably and to a comparable degree as lesions in the anterior spine, with the exception of lesions in the cervical spine and soft tissues. Fifth, the addition of a simple dichotomous method for scoring posterior element inflammation substantially enhanced the discrimination provided by the SPARCC MRI score for vertebral body inflammation.

The assessment of spinal inflammation in patients with AS by MRI has, to date, focused on the evaluation of lesions in the anterior spine because they are readily visible, whereas lesions in the lateral and posterior structures are more difficult to interpret due to their small size and complex anatomy. In addition, standard protocols for routine clinical MRI do not visualize the transverse processes at all unless transverse sequences are performed to evaluate anterior or spinal canal pathology, and then only a limited number of processes can be seen. Nevertheless, this work and previous work from our group describing systematic evaluation of lesions in the lateral structures shows that these lesions are at least as frequent as those in the anterior spine (11). The lateral and posterior structures must therefore be scrutinized when patients with established spondylarthritis are referred for either diagnostic evaluation or assessment of disease activity status. This may not be possible in routine imaging of the spine using MRI because this assessment is most frequently conducted for orthopedic or neurologic indications in which neuroforamina and lesions of the disc and spinal canal are the primary focus. Lesions in the lateral structures require additional slices that extend well beyond the pedicles, a problem that is greatly exaggerated if there is even a minor degree of scoliosis present.

One limitation of our study is that all patients had active AS that was refractory to treatment with nonsteroidal antiinflammatory drug therapy. We did not include patients early in their disease course, when the frequency and distribution of lesions may differ from those in established disease. A second issue is the possibility of underscoring of the transverse process lesions. Prior to this study, our diagnostic imaging protocol specified that scans must extend beyond the pedicles on either side, and the number of slices ranged from 11–15. Since this study, we have amended our spondylarthritis diagnostic imaging protocol to require a minimum of 16 sagittal slices and inclusion of transverse processes bilaterally at every level.

The higher frequency of lesions in the thoracic spine is consistent with the predilection for the thoracic spine that has been reported previously when anterior lesions were systematically evaluated (12). In addition, our work supports the view that the thoracolumbar junction is preferentially affected. Although the finding of lesions in spinous processes is consistent with enthesitis at the attachment of spinal ligaments, transverse process edema was observed at least as often and may be related to either enthesitis or inflammation in the costotransverse articulations. Inflammation in the facet joints is consistent with their involvement in this disease. However, the facet joints and pedicles are depicted in the lateral sagittal slices that also depict the lateral portion of the vertebral body. Inflammation at the latter site may reflect a vertebral corner lesion focused on the lateral lip of the vertebral body, a spondylodiscitis focused on the lateral portion of the vertebral endplate, or costovertebral inflammation in the thoracic spine. The association between lesions in the vertebral body and lesions in the pedicles and facet joints may therefore be interpreted in two respects. Inflammation may originate in the vertebral body and then extend posteriorly, or the converse may apply, and the readers were often unable to determine the primary source of inflammation by its appearance on MRI.

The 2 observers achieved good to excellent reliability in the detection of lesions. This was achieved with minimal training and was facilitated by a systematic approach to the evaluation of posterior structures using an electronic, online scoring module that was specifically designed for this purpose. Reliability was particularly good for assessment of lesions in the thoracic spine, where the reliability of change scores after treatment with anti-TNF agents contributes to the high discrimination that was noted when the inflammation in posterior elements was added to the SPARCC scores for vertebral body inflammation. Reliability was less good for the assessment of lesions in the cervical spine because cervical vertebrae appear small on the large field of view, which is necessary for depicting the entire spine in 2 halves, the cervicothoracic and thoracolumbar portions. Consequently, resolution is poor and the data provided in this report should be regarded as conservative estimates. Detection of soft tissue lesions was also less reliable, principally due to the variable but frequent presence of multiple small veins that appear bright on the T2-weighted sequences, obfuscating the otherwise dark background.

In addition to good reliability, our data also show that a simple method of scoring inflammatory lesions in the posterior structures on a dichotomous basis is also responsive to treatment intervention and substantially enhances the discrimination of the SPARCC spinal inflammation score, which is based solely on vertebral body inflammation. Moreover, the very high discrimination observed with this extended SPARCC score may prove to be particularly useful for proof-of-concept studies of new therapeutic agents.

The pathophysiologic and prognostic significance of these lesions requires further study. One report about patients with long-standing AS undergoing facet joint resection and spinal fusion has shown that lesions on MRI are associated with histopathologic features of inflammation, especially neovascularization and infiltration with macrophages and osteoclasts (13). However, MRI lacked sensitivity, detecting lesions in only 3 of 8 patients with histopathologic evidence of inflammation. It would be important to know whether lesions on MRI predict the development of radiographic abnormalities, especially fusion of facet joints, costovertebral joints, or costotransverse joints, because this may lead to major functional impairment. Such an association would increase the importance of MRI in guiding treatment.

In conclusion, we have conducted the first systematic assessment to our knowledge of posterior structures in patients with AS using MRI, and have shown that inflammatory lesions occur in almost all patients with established and active AS in the posterior structures almost as frequently as in the anterior spine. They are most frequently found in the thoracic spine and at transverse and/or spinous processes, where they are reliably detected and are ameliorated with anti-TNFα therapy. Assessment of posterior structures should constitute a requirement for routine diagnostic MRI assessment of the spine in patients with established AS or suspected of having AS. In addition, the additional quantification of inflammation in posterior structures using a simple dichotomous scoring method substantially enhances the discrimination of spinal MRI in the assessment of new therapeutics. Our data therefore carry important implications for both diagnosis and clinical trials research.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Maksymowych had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Maksymowych, Dhillon, Crowther, Lambert.

Acquisition of data. Maksymowych, Dhillon, Crowther, Lambert.

Analysis and interpretation of data. Maksymowych, Conner-Spady, Lambert.