The diagnostic utility of magnetic resonance imaging in spondylarthritis: An international multicenter evaluation of one hundred eighty-seven subjects

Authors


Abstract

Objective

To systematically assess the diagnostic utility of magnetic resonance imaging (MRI) to differentiate patients with spondylarthritis (SpA) from patients with nonspecific back pain and healthy volunteers, using a standardized evaluation of MR images of the sacroiliac joints.

Methods

Five readers blinded to the patients and diagnoses independently assessed MRI scans (T1-weighted and STIR sequences) of the sacroiliac joints obtained from 187 subjects: 75 patients with ankylosing spondylitis (AS; symptom duration ≤10 years), 27 patients with preradiographic inflammatory back pain (IBP; mean symptom duration 29 months), 26 patients with nonspecific back pain, and 59 healthy control subjects; all participants were age 45 years or younger. Bone marrow edema, fat infiltration, erosion, and ankylosis were recorded according to standardized definitions using an online data entry system. We calculated sensitivity, specificity, and positive and negative likelihood ratios (LRs) for the diagnosis of SpA based on global assessment of the MRI scans.

Results

Diagnostic utility was high for all 5 readers, both for patients with AS (sensitivity 0.90, specificity 0.97, positive LR 44.6) and for patients with preradiographic IBP (sensitivity 0.51, specificity 0.97, positive LR 26.0). Diagnostic utility based solely on detection of bone marrow edema enhanced sensitivity (67%) for patients with IBP but reduced specificity (88%); detection of erosions in addition to bone marrow edema further enhanced sensitivity (81%) without changing specificity. A single lesion of the sacroiliac joint on MRI was observed in up to 27% of control subjects.

Conclusion

This systematic and standardized evaluation of sacroiliac joints in patients with SpA showed that MRI has much greater diagnostic utility than has been documented previously. We present for the first time a data-driven definition of MRI-visualized positivity for SpA.

Diagnosing early spondylarthritis (SpA) in young patients presenting with symptoms of inflammatory back pain (IBP) and normal findings on plain radiographs of the sacroiliac (SI) joints remains a challenge in routine practice. Radiography detects postinflammatory structural changes in the subchondral bone of the SI joints, but changes may become evident only after a symptom duration of several years (1, 2). Magnetic resonance imaging (MRI) is capable of detecting inflammatory changes in bone marrow and soft tissue and is now widely accepted as the most sensitive imaging modality for detecting sacroiliitis, the hallmark of SpA. Therefore, MRI may be able to detect abnormalities in the SI joints prior to detection using radiography. A systematic literature review in 2004 (3) addressed those studies that focused on the diagnostic utility of MRI in patients with either established ankylosing spondylitis (AS) or IBP and concluded that the overall sensitivity and specificity of MRI for SpA was 0.90, with a positive likelihood ratio (LR) of 9.0. This estimate of diagnostic utility has since been incorporated into diagnostic algorithms for diagnosing SpA in clinical practice, which rely on estimates of the degree to which positive findings by MRI increase the pretest probability of SpA.

There are several reasons, however, why this estimate of diagnostic utility requires reappraisal. Most studies included in the systematic review lacked age- and sex-matched controls and used dynamic (not standard) contrast-enhanced MRI, which is costly, unreliable, and not routinely used in clinical practice (4–8). Moreover, there is increasing acceptance of the necessity for including certain methodologic aspects of study design such as standardization of technique for acquiring and reading MR images, consensus definitions of abnormalities visible on MR images, and standardized calibration/training of readers at different sites. These requirements are aimed at ensuring external validity of study conclusions.

The recently published Assessment of SpondyloArthritis international Society (ASAS) classification criteria for axial spondylarthritis (9) include, for the first time, an MR image demonstrating sacroiliitis as an acceptable imaging criterion indicative of SpA but additionally require the presence of at least 1 clinical feature of SpA. An ASAS working group has also proposed a definition of an MR image positive for sacroiliitis according to consensus opinion that is based entirely on the presence of bone marrow edema on the STIR sequence or osteitis on gadolinium-enhanced T1-weighted images (10). Structural abnormalities that may be observed on T1-weighted sequences are not included in this definition.

In this report, we describe a multicenter evaluation of the diagnostic utility of MRI in patients with SpA that includes age- and sex-matched control subjects, uses MRI sequences that are widely used in clinical practice, and is based on a standardized methodology for the diagnostic evaluation of the SI joint by MRI, as follows: implementation of standardized definitions of active inflammatory and structural lesions of the SI joint on MRI as defined by the Canada–Denmark MRI Working Group (11, 12); development of a reference MR image set of an SI joint by consensus among study investigators based on these definitions (www.arthritisdoctor.ca); calibration of readers using an online training module developed by the Spondyloarthritis Research Consortium of Canada (SPARCC) (ref.13 and www.arthritisdoctor.ca) followed by videoteleconferences; and development of a customized online data entry module based on a standardized approach to recording abnormalities in the SI joint (www.arthritisimaging.ca).

The study was called MORPHO, because the morphologic appearance of bright bone marrow edema on STIR images involving both sides of each sacroiliac joint resembles the pattern of light reflection on opened wings of Morpho butterflies (Morpho sp.) that are found in South American rainforests.

PATIENTS AND METHODS

Subjects.

Patients with SpA.

One hundred two patients with SpA, ages ≤45 years, were recruited at the rheumatology outpatient clinics of 2 university hospitals in Edmonton and Zurich. Of these patients, 75 met the modified New York classification criteria for AS (14) and had a disease duration of ≤10 years. Nineteen patients with IBP (all recruited from Balgrist University Hospital, Zurich) not fulfilling the modified New York criteria for radiographic AS did meet the Berlin criteria for IBP (15) and additionally showed ≥1 clinical or laboratory feature of SpA. In 8 patients with IBP (all from the University of Alberta) not meeting the modified New York criteria, disease was defined according to the criteria described by Calin and colleagues (16) and by expert opinion. Among the 27 patients with IBP, the mean symptom duration was 29 months. Patients with SpA who had received ongoing or previous treatment with tumor necrosis factor α inhibitors were not eligible for this study. Plain pelvic radiographs obtained from patients in both SpA groups were independently assessed and categorized according to the modified New York criteria (14), by 2 readers at each site.

Control groups.

The first control group comprised 26 patients ages ≤45 years, all of whom had nonspecific back pain. Nonspecific back pain was defined based on clinical grounds and according to plain radiographs of the SI joint. A second control group comprised 59 healthy volunteers (ages ≤45 years) from the staffs of both university hospitals that were involved in the recruitment of patients with SpA. The eligibility of these volunteers was defined using the Nordic questionnaire (17); volunteers with symptoms compatible with IBP or clinical features suggestive of SpA (e.g., psoriasis of the skin or chronic inflammatory bowel disease) were excluded.

The study protocol was approved by the local ethics committees; the patients and the healthy participants all gave written informed consent.

MRI protocol.

Scans from both centers included coronal T1-weighted turbo spin-echo and STIR sequences angled parallel to the SI joint. The scan parameters for all sequences were as follows: 15–19 slices, 4-mm slice thickness, 0.4-mm interslice gap, and field of view 280–300 mm. Parameters for the T1-weighted sequence were as follows: repetition time (TR) 423–450 msec, echo time (TE) 12–13 msec, echo train length (ETL) 3, and matrix 512 × 256 pixels. For the STIR sequence, the parameters were TR 3,700–4,930 msec, inversion time 145–150 msec, TE 50–69 msec, ETL 7–9, and matrix 256–384 × 256 pixels. These are the usual sequences and scan parameters for routine MRI evaluation of patients with SpA in the involved institutions.

Standardized assessment of MR images.

STIR and T1-weighted sequences of semicoronal MR scans of the SI joint were read and scored independently by 5 readers (2 radiologists [JH and RGL] and 3 rheumatologists [MØ, UW, and WPM]) who were blinded to patient demographics and diagnosis. The films were evaluated in random order on electronic work stations in the institution of each reader. The following steps were performed in chronological order.

Standardized definitions of lesions on MRI scans

We adopted standardized definitions of active inflammatory and structural lesions of the SI joint on MRI, which were developed by the Canada–Denmark MRI Working Group (11, 12). We focused on 4 types of lesions identified by MRI: bone marrow edema on STIR images, and joint erosion, marrow fat infiltration, and ankylosis on T1-weighted images (Figures 1A–D). Bone sclerosis and abnormalities of the synovial cavity were not addressed because of poor reproducibility in prereadout exercises.

Figure 1.

Four types of acute and structural lesions of the sacroiliac (SI) joint as demonstrated by magnetic resonance imaging. A, Bone marrow edema (arrows) in all quadrants of both SI joints of a 28-year-old HLA–B27–positive female patient with inflammatory back pain and a symptom duration of 15 months, on STIR sequences. Bone marrow edema is defined as an increase in bone marrow signal in the SI joint on STIR images; the center of the sacrum at the same craniocaudal level is used as the primary reference for normal bone marrow signal. B, Erosion at the sacral side of the right SI joint (arrow) of a 25-year-old HLA–B27–positive male patient with ankylosing spondylitis (AS) with a symptom duration of 24 months, on T1-weighted spin-echo (T1SE) sequences. An erosion is defined as the full-thickness loss of dark appearance of either iliac or sacral cortical bone of the SI joint and change in normal bright appearance of adjacent bone marrow on T1-weighted images; adjacent bone marrow demonstrates altered signal intensity on T1-weighted images as compared with normal iliac or sacral marrow on the same slice at the same craniocaudal level. The lower iliac portion of the right SI joint shows an erosion at the threshold of definition (broken arrow), while sclerosis (dotted arrow) is seen in the upper iliac half of the right SI joint (open arrowheads point to the marrow signal medial to the dark area that is similar to adjacent iliac marrow). The left SI joint displays sclerosis and definite erosion in the left iliac bone (solid arrowheads) and, as a consequence, pseudo-widening of joint space; loss of adjacent marrow signal is visible compared with normal iliac marrow. C, Fat infiltration predominantly on the sacral side of both SI joints (arrows) in a 28-year-old HLA–B27–positive male patient with AS with a symptom duration of 7 years, on T1-weighted spin-echo sequences. Fat infiltration is defined as a focal increased signal in bone marrow on T1-weighted images. D, Ankylosis (arrows) in the distal part of both SI joints of a 42-year-old HLA–B27–negative male patient with AS with a symptom duration of 8 years, by T1-weighted spin-echo sequences. Ankylosis is defined as bright signal on T1-weighted images extending across the SI joint.

Reference set of SI joint MR images

We developed a reference SI joint MR image set by consensus among study investigators, based on the definitions for these 4 types of lesion (ref.13 and www.arthritisdoctor.ca). Four videoteleconference sessions involving the 3 university centers focused on the application of the standardized definitions of SI joint lesions on MRI and served to calibrate the reader team by conducting 2 test-reading exercises on SI joint scans displaying acute and structural inflammatory lesions. The reference image set served as a benchmark at the time of reading the study MR images and included lesions considered to be at the threshold of detection, to facilitate calibration and the assignment of questionable lesions.

Standardized assessment of MRI-visualized lesions of the SI joint

We standardized the approach to assessing MR images of the SI joint by adopting the methodology outlined in the SPARCC online training module (13, 18). This training module has been validated by demonstrating that inexperienced rheumatology fellows can achieve reliability of detection of bone marrow edema comparable with that recorded by experienced SPARCC readers after reviewing the training module (13). Principal features of the module include systematic assessment of the SI joint from anterior to posterior. At least 1 SI joint must be a minimum of 1 cm in vertical height in order to score that image; once an SI joint is at least 3 cm in vertical height, it is divided at the midpoint into 4 equal upper and lower sacral and iliac quadrants. At the posterior aspect of the SI joint, there is a natural division of the joint into upper and lower quadrants by intervening fat and fibrous tissue.

Online data entry module

We developed a customized online data entry module for recording MRI findings based on a standardized approach to recording abnormalities in the SI joint. The module has 2 sections, the first of which contains 3 questions that address global assessment of each scan, as follows: 1) “This MRI scan confirms the presence of SpA (agree/disagree)”; 2) “Your conclusion is based on which MRI sequence (STIR, T1-weighted turbo spin-echo, both sequences)”; and 3) “What is the primary MRI feature on which your diagnosis of SpA is based (bone marrow edema, fat infiltration, bone erosion, ankylosis, not applicable because SpA is not present).” The second section of the Web-based data entry module consists of a detailed recording section in which the SI joint is represented as a schematic with 4 quadrants (upper and lower ilium, upper and lower sacrum). Each lesion, except ankylosis, is recorded as being present/absent on a dichotomous basis in each quadrant. Ankylosis is recorded in each half of the joint (upper and/or lower) (Figure 1).

Statistical analysis.

The distribution of the 4 MRI lesions in patients and control subjects was analyzed descriptively according to single readers, mean (range) values for all 5 readers, and concordant data recorded by at least 2 readers. The ASAS-proposed definition of a positive MRI result based on consensus opinion stipulates the requirement for 2 bone marrow edema lesions on the same slice or, if only a single bone marrow edema lesion is evident, it must be present on 2 consecutive slices (10). We considered this definition as being met in our data entry module in 2 circumstances. The first would be when a score of at least 2 for bone marrow edema was recorded on the same slice, indicating 2 bone marrow edema lesions in 2 distinct SI joint quadrants or a single lesion extending across 2 SI joint quadrants. The second circumstance would also be a score of at least 2 for bone marrow edema, but this would be attributable to a score of 1 indicating a single bone marrow edema lesion being recorded on 2 consecutive slices in the same SI joint quadrant. We adopted a similar approach to descriptive data for erosions and fat infiltration.

The diagnostic utility of MRI for SpA according to global evaluation of both T1-weighted and STIR MRI scans was determined by calculating the sensitivity, specificity, positive and negative LRs for individual reader data, concordant data according to at least 2 readers, and for data recorded concordantly by all 3 rheumatologists and both radiologists. These analyses were repeated for the proposed ASAS definition of a positive result of MRI. We also formulated a data-driven proposal for a positive MRI result after recording a high frequency and specificity of erosions in patients with preradiographic SpA and after noting that the diagnostic utility of MRI according to the ASAS definition was less than that recorded by global assessment. We postulated that global assessment captured additional diagnostic information beyond that provided by assessment of bone marrow edema, and that this information would largely be attributable to the presence of erosions. We therefore proposed an alternate definition of a positive MRI result, termed the MORPHO proposal, which defines SpA as being present on MRI if any of the following 3 criteria are met: 1) bone marrow edema in at least 2 SI joint quadrants in the same slice or a single SI joint quadrant in 2 consecutive slices (according to the ASAS definition); 2) erosion in at least 2 SI joint quadrants in the same slice or a single SI joint quadrant in 2 consecutive slices; and 3) bone marrow edema and erosion in any SI joint quadrant though not necessarily in the same quadrant. The diagnostic utility of this definition was tested by calculating sensitivity, specificity, positive and negative LRs for individual reader data, concordant data according to at least 2 readers, and data recorded by all rheumatologists and both radiologists.

RESULTS

Descriptive analysis.

Demographics of patients with SpA and control groups.

Among the 75 patients meeting the modified New York criteria, the mean ± SD age was 31.1 ± 6.2 years, the mean ± SD disease duration was 6.1 ± 2.8 years, and 72% of the patients were male. The mean ± SD Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) (19) score was 4.4 ± 1.9, and the mean ± SD Bath Ankylosing Spondylitis Functional Index (BASFI) (20) score was 2.9 ± 2.5; both of these scoring systems are based on a scale of 0–10. The mean ± SD C-reactive protein (CRP) level was 10.5 ± 18.4 mg/liter. Fifty-nine (83%) of 71 patients with AS who underwent HLA–B27 testing had positive results. The 27 patients with IBP (67% of whom were male) had a mean ± SD age of 29 ± 6.8 years and a mean ± SD symptom duration of 29 ± 26 months; their mean ± SD BASDAI and BASFI scores were 4.2 ± 2.1 and 2.6 ± 2.2, respectively, and their mean ± SD CRP levels were 13.8 ± 30.8 mg/liter. Twenty-three (92%) of 25 patients with IBP who underwent testing for HLA–B27 had positive results. Among the 59 healthy control subjects, the mean ± SD age was 31.0 ± 6.2 years, and 63% were male. The mean ± SD age of the 26 patients with nonspecific back pain was 33.8 ± 7.9 years, and 58% were male.

Frequency of specific MRI abnormalities.

The number and percent of patients with SpA and control subjects with MRI lesions that were recorded concordantly by any 2 readers and the mean number (range and percent) recorded by all 5 readers for the 4 categories of MRI lesions are shown in Table 1. Bone marrow edema, erosion, and fat infiltration were recorded in both patients and control subjects, while ankylosis was recorded only in patients with SpA and those with nonspecific back pain. Bone marrow edema was the most common inflammatory lesion in the group with IBP, followed by erosion and fat infiltration. In contrast, erosion and fat infiltration were recorded more frequently than bone marrow edema in the group with AS. Bone marrow edema meeting the proposed ASAS criteria for a positive MRI result was recorded concordantly in 66.7% (18 of 27) and 85.3% (64 of 75) of patients with IBP and patients with AS, respectively, but it was also recorded concordantly in 23.1% of patients with nonspecific back pain and in 6.8% of healthy control subjects. A single bone marrow edema lesion was observed in 26.9% of the nonspecific back pain patients and in 22% of the healthy control subjects. When the ASAS operational definition was applied to fat infiltration, fat infiltration was recorded concordantly as being present in at least 2 SI joint quadrants on the same slice or in 1 SI joint quadrant on 2 consecutive slices in 90.7%, 37%, 15.4%, and 13.6% of the patients with AS, those with IBP, patients with nonspecific back pain, and healthy control subjects, respectively. Applying the ASAS operational definition to erosions, this was recorded concordantly in 90.7% and 48.1% of patients with AS and those with IBP, respectively, but in only 3.8% of patients with nonspecific back pain and 1.7% of healthy control subjects.

Table 1. Frequency of specific MRI abnormalities in SpA patients and controls, as recorded concordantly by ≥2 readers and according to the mean for all 5 readers*
GroupBone marrow edemaErosionFat infiltrationAnkylosis
  • *

    Values are the number (%) of abnormalities recorded concordantly by ≥2 readers/mean number of abnormalities (% [range]) recorded by all 5 readers. SpA = spondylarthritis; AS = ankylosing spondylitis; NA = not applicable (ankylosis was recorded in each half of the joint, not per quadrant); IBP = inflammatory back pain; NSBP = nonspecific back pain.

  • Assessment of SpondyloArthritis international Society (ASAS) proposal for magnetic resonance imaging (MRI) positive for abnormalities (10): ≥2 bone marrow edema (BME) lesions in 2 distinct sacroiliac joint (SIJ) quadrants on the same slice or ≥1 BME lesion extending across 2 SIJ quadrants or ≥1 BME lesion recorded on 2 consecutive slices in the same SIJ quadrant.

  • ASAS operational definition adapted to erosions and fat infiltration (presence of MRI lesions in at least 2 SIJ quadrants in the same slice or in a single SIJ quadrant on 2 consecutive slices).

AS patients (n = 75)    
 ≥1 SIJ quadrant68 (90.7)/62.8 (83.7 [59–68])70 (93.3)/59.4 (79.2 [42–71])70 (93.3)/65.2 (86.9 [64–67])20 (26.7)/18.4 (24.5 [11–27])
 ≥2 SIJ quadrants65 (86.7)/60.4 (80.5 [57–67])68 (90.7)/56.6 (75.5 [37–68])68 (90.7)/62.2 (82.9 [61–66])14 (18.7)/14.2 (18.9 [10–19])
 ASAS proposal64 (85.3)/59.8 (79.7 [57–66])68 (90.7)/56 (74.7 [37–68])68 (90.7)/62 (82.7 [60–66])NA
 ≥3 SIJ quadrants62 (82.7)/58.0 (77.3 [54–66])63 (84.0)/52.2 (69.6 [32–68])66 (88.0)/58.8 (78.4 [55–64])13 (17.3)/12.6 (16.8 [9–18])
IBP patients (n = 27)    
 ≥1 SIJ quadrant21 (77.8)/18.6 (68.9 [16–21])16 (59.3)/13.0 (48.1 [8–20])12 (44.4)/9.6 (35.6 [7–12])0 (0)/0.2 (0.007 [0–1])
 ≥2 SIJ quadrants18 (66.7)/16.4 (60.7 [14–19])16 (59.3)/11.6 (43.0 [6–19])10 (37.0)/8.6 (31.9 [6–11])0 (0)/0 (0 [0])
 ASAS proposal18 (66.7)/16.2 (60.0 [14–18])13 (48.1)/10.4 (38.5 [5–18])10 (37.0)/8.4 (31.1 [6–11])NA
 ≥3 SIJ quadrants16 (59.3)/15.4 (57.0 [14–17])11 (40.7)/8.2 (30.4 [4–14])9 (33.3)/7.2 (26.7 [5–9])0 (0)/0 (0 [0])
NSBP patients (n = 26)    
 ≥1 SIJ quadrant7(26.9)/6.2 (23.8 [3–12])2 (7.7)/2.8 (10.8 [0–6])5 (19.2)/4.6 (17.7 [2–7])0 (0)/0.4 (0.02 [0–1])
 ≥2 SIJ quadrants6 (23.1)/4.4 (16.9 [2–9])2 (7.7)/1.6 (6.2 [0–3])5 (19.2)/3.8 (14.6 [2–6])0 (0)/0.2 (0.008 [0–1])
 ASAS proposal6 (23.1)/4.2 (16.2 [2–8])1 (3.8)/1.2 (4.6 [0–2])4 (15.4)/3.6 (13.8 [2–6])NA
 ≥3 SIJ quadrants5 (19.2)/3.4 (13.1 [2–5])1 (3.8)/0.8 (3.1 [0–2])2 (7.7)/2.4 (9.2 [2–3])0 (0)/0.2 (0.008 [0–1])
Healthy controls (n = 59)    
 ≥1 SIJ quadrant13 (22.0)/11 (18.6 [4–19])5 (8.5)/5.8 (9.8 [1–13])14 (23.7)/10.2 (17.3 [6–19])0 (0)/0 (0 [0])
 ≥2 SIJ quadrants7 (11.9)/5.6 (9.5 [3–12])2 (3.4)/3 (5.1 [1–10])9 (15.3)/7.2 (12.2 [4–14])0 (0)/0 (0 [0])
 ASAS proposal4 (6.8)/4.6 (7.8 [2–9])1 (1.7)/2 (3.4 [0–7])8 (13.6)/7.2 (12.2 [4–14])NA
 ≥3 SIJ quadrants5 (8.5)/3.6 (6.1 [1–8])1 (1.7)/1 (1.7 [0–3])8 (13.6)/5.4 (9.2 [3–11])0 (0)/0 (0 [0])

Frequency of a diagnosis of SpA in the 4 groups based on results of the global assessment of SI joint MRI.

The number (percent) of patients in whom SpA was diagnosed concordantly by ≥2 readers according to the global evaluation of both MRI sequences was 74 (99%) of 75 in the AS group and 14 (52%) of 27 in the IBP group. The 13 patients with IBP in whom SpA was not diagnosed demonstrated few SI joint quadrants with bone marrow edema, erosion, and fat infiltration (11 patients) or no lesions on MRI at all (2 patients). One patient with AS in whom SpA was not diagnosed was reported as having no bone marrow edema by any reader, although erosions were recorded in up to 5 SI joint quadrants, and fat infiltration was recorded in up to 7 SI joint quadrants. In 2 (8%) of 26 patients with nonspecific back pain, SpA was diagnosed by 2 readers each, primarily on the basis of bone marrow edema in 1 patient and both bone marrow edema and erosions in the second patient. SpA was diagnosed in 3 (5%) of 59 healthy volunteers (in 1 control subject by 3 readers and in 2 control subjects by 2 readers, respectively). The allocation of the 3 healthy volunteers to the SpA group was mainly based on the presence of bone marrow edema.

Concordance and discordance for the diagnosis of SpA according to global assessment by MRI.

The concordance for the overall assessment between all 5 readers, the 3 rheumatologists and both radiologists responding to the question “This MRI scan confirms the presence of SpA” is shown in Table 2. Concordance was defined as all readers in a group being in agreement that the MRI scan indicated a diagnosis of SpA or no SpA. Readers were defined as being discordant for either SpA or no SpA depending on whether the majority considered the MR image as indicating SpA or no SpA, respectively. Because there were only 2 radiologists, this definition for discordance could not be applied to the diagnoses reported by them. Concordance for the diagnosis of SpA by MRI in the IBP group was 85% for all 5 readers, 89% for the 3 rheumatologists, and 96% for both radiologists. Concordance for the absence of SpA in the 2 control groups was high, with 92–96% concordance for the nonspecific back pain control subjects and 93–97% concordance for the healthy volunteers.

Table 2. Concordance for the diagnosis of SpA according to global assessment by MRI*
Group, readersAgreement for SpAAgreement for no SpA
ConcordantDiscordantConcordantDiscordant
  • *

    Values are the number (%). Concordance was achieved when all readers of a reader group reported that the magnetic resonance image (MRI) indicated spondylarthritis or no spondylarthritis. Discordance was reported when the reports of the readers were discordant for either spondylarthritis or no spondylarthritis. AS = ankylosing spondylitis; NA = not applicable (the definition for discordance could not be applied to the diagnoses reported because there were only 2 radiologists); IBP = inflammatory back pain.

  • In 2 patients with nonspecific back pain (NSPB), NSPB was diagnosed as spondylarthritis (SpA) by 2 of 5 readers each.

  • In 1 healthy control subject, SpA was diagnosed by 3 of 5 readers; in 2 control subjects, SpA was diagnosed by 2 of 5 readers each.

AS patients (n = 75)    
 All 5 readers51 (68)19 (25)0 (0)5 (7)
 All 3 rheumatologists56 (75)13 (17)2 (3)4 (5)
 Both radiologists65 (87)10 (13)0 (0)NA
IBP patients (n = 27)    
 All 5 readers12 (44)2 (7)11 (41)2 (7)
 All 3 rheumatologists13 (48)0 (0)11 (41)3 (11)
 Both radiologists13 (48)1 (4)13 (48)NA
NSBP patients (n = 26)    
 All 5 readers0 (0)0 (0)24 (92)2 (8)
 All 3 rheumatologists0 (0)1 (4)24 (92)1 (4)
 Both radiologists0 (0)NA25 (96)1 (4)
Healthy controls (n = 59)    
 All 5 readers0 (0)1 (2)55 (93)3 (5)
 All 3 rheumatologists1 (2)1 (2)56 (95)1 (2)
 Both radiologists0 (0)NA57 (97)2 (3)

Diagnostic utility of MRI for SpA.

Diagnostic utility of MRI based on global assessment.

Table 3 displays the mean (range) sensitivity, specificity, and positive and negative LRs for all 5 readers for the diagnosis of SpA following global assessment by MRI as well as diagnostic assignments recorded concordantly by all 3 rheumatologists, by both radiologists, and by ≥2 of the 5 readers. Diagnostic utility was, as expected, very high in patients with established AS (the mean positive LR and mean negative LR for all 5 readers were 44.6 and 0.10, respectively). Specificity was consistently high among all readers (range 94–99%), while sensitivity was somewhat lower for rheumatologist readers. The diagnostic utility of MRI for IBP was also high (the mean positive LR and negative LR for all 5 readers were 26.0 and 0.50, respectively), with few differences between rheumatologists and radiologists.

Table 3. Diagnostic utility of MRI for SpA as determined by global assessment of MR images by individual readers and according to diagnostic assignments recorded concordantly*
Group, readersSensitivitySpecificityPositive LRNegative LR
  • *

    Except where indicated otherwise, values are the diagnostic assignments recorded concordantly by all 3 rheumatologists, by both radiologists, and by at least 2 of the 5 readers. MRI = magnetic resonance imaging; SpA = spondylarthritis; LR = likelihood ratio; AS = ankylosing spondylitis; NSBP = nonspecific back pain; HC = healthy control; NC = not calculable (specificity = 1.0); IBP = inflammatory back pain.

AS versus NSBP and HC combined    
 Mean (range) of 5 readers0.90 (0.83–0.97)0.97 (0.94–0.99)44.6 (16.3–72.5)0.10 (0.01–0.18)
 Concordant assignments    
  3 rheumatologists0.750.9967.90.26
  2 radiologists0.871.0NC0.13
  ≥ any 2 readers0.990.9518.60.01
AS versus NSBP    
 Mean (range) of 5 readers0.90 (0.83–0.99)0.97 (0.92–1.0)NC (12.5–NC)0.10 (0.01–0.17)
 Concordant assignments    
  3 rheumatologists0.751.0NC0.25
  2 radiologists0.871.0NC0.13
  ≥ any 2 readers0.990.9212.80.01
AS versus HC    
 Mean (range) of 5 readers0.90 (0.83–0.99)0.97 (0.95–0.98)40.3 (18.8–58.0)0.10 (0.01–0.17)
 Concordant assignments    
  3 rheumatologists0.750.9953.40.26
  2 radiologists0.871.0NC0.13
  ≥ any 2 readers0.990.9623.00.01
IBP versus NSBP and HC combined    
 Mean (range) of 5 readers0.51 (0.48–0.52)0.97 (0.94–0.99)26.0 (8.8–43.3)0.50 (0.49–0.53)
 Concordant assignments    
  3 rheumatologists0.480.9943.60.53
  2 radiologists0.481.0NC0.52
  ≥ any 2 readers0.520.959.80.51
IBP versus NSBP    
 Mean (range) of 5 readers0.51 (0.48–0.52)0.97 (0.92–1.0)NC (6.7–NC)0.50 (0.48–0.54)
 Concordant assignments    
  3 rheumatologists0.481.0NC0.52
  2 radiologists0.481.0NC0.52
  ≥ any 2 readers0.520.926.70.52
IBP versus HC    
 Mean (range) of 5 readers0.51 (0.48–0.52)0.97 (0.95–0.98)23.0 (10.2–30.5)0.50 (0.49–0.53)
 Concordant assignments    
  3 rheumatologists0.480.9934.30.53
  2 radiologists0.481.0NC0.52
  ≥ any 2 readers0.520.9612.10.50

Diagnostic utility of MRI based on the ASAS definition of a positive MRI.

Table 4 compares the diagnostic utility of MRI as determined by global evaluation of all scans versus the ASAS definition of a positive MRI result. While sensitivity increased for all 5 readers when using the ASAS definition, there was a decrease in specificity when compared with global evaluation of MRI. The diagnostic utility of the ASAS definition was therefore less than that determined by global assessment for all 5 readers.

Table 4. Diagnostic utility of MRI by global assessment of all scans compared with the ASAS and MORPHO definitions of a positive MRI in patients with IBP*
 Reader 1Reader 2Reader 3Reader 4Reader 5Any 2 readers
  • *

    The Assessment of SpondyloArthritis international Society (ASAS) proposal for a positive magnetic resonance image (MRI) is ≥2 bone marrow edema (BME) lesions in 2 distinct sacroiliac joint (SIJ) quadrants on the same slice or ≥1 BME lesion extending across 2 SIJ quadrants or ≥1 BME lesion recorded on 2 consecutive slices in the same SIJ quadrant. The MORPHO proposal for a positive MRI is fulfillment of any of the following 3 criteria: BME according to the ASAS definition, erosion in ≥2 SIJ quadrants in the same slice or a single SIJ quadrant in 2 consecutive slices, or BME and erosion in any SIJ quadrant though not necessarily in the same quadrant. IBP = inflammatory back pain; LR = likelihood ratio.

Diagnosis by overall assessment      
 Sensitivity0.480.520.520.520.520.52
 Specificity0.980.970.940.990.990.95
 Positive LR20.014.88.843.443.39.8
 Negative LR0.50.50.50.50.50.5
Diagnosis by ASAS proposal      
 Sensitivity0.520.590.670.560.670.67
 Specificity0.930.920.850.940.890.88
 Positive LR7.37.34.58.86.05.7
 Negative LR0.50.40.40.50.40.4
Diagnosis by MORPHO proposal      
 Sensitivity0.740.630.780.700.740.81
 Specificity0.860.910.850.920.890.88
 Positive LR5.37.05.28.86.76.9
 Negative LR0.30.30.30.30.30.2

Diagnostic utility of MRI based on the MORPHO definition of a positive MRI.

The inclusion of erosions improved sensitivity (81%) as compared with the ASAS definition (67%) for all 5 readers, with minimal change in specificity, so that overall diagnostic utility was better (Table 4). Although the positive LR (6.9) was still not quite as good as global assessment of MRI (positive LR = 9.8), the negative LR was better (0.2 versus 0.5).

DISCUSSION

This international multireader MRI standardization, calibration, and reading exercise involving 187 patients with SpA and age- and sex-matched control subjects demonstrated several findings that carry major clinical implications for the use of MRI in SpA. First, the adoption of a systematic and standardized approach to assessment of the SI joint shows that MRI has much greater diagnostic utility than has been documented previously. Second, structural changes occur early in the disease course and are evident in the SI joint even in patients with a symptom duration of <24 months. Third, bone marrow edema and the presence of fat infiltration in >1 SI joint quadrant or extending into consecutive slices is relatively nonspecific and may occur even in healthy, asymptomatic individuals. Fourth, erosions are relatively specific for SpA. Fifth, limiting assessment to bone marrow edema may fail to capture significant diagnostic information available from MR sequences obtained during routine screening.

Our study protocol incorporated several issues relevant to rheumatology practice. The inclusion criteria focused on a young population representing the demographic in which SpA develops and also on patients with recent-onset SpA who may represent a challenge for early diagnosis. Young persons with nonspecific back pain and healthy individuals constitute essential control participants for studies of diagnostic utility in patients with early SpA in whom unequivocal radiographic sacroiliitis has yet to develop and therefore reflect a frequent diagnostic challenge in daily practice. MRI protocols using T1-weighted and STIR sequences to evaluate the SI joints of patients with clinically suspected SpA are widely used in clinical practice; 2 recent studies showed that additional expensive and time-consuming contrast-enhanced MRI sequences do not improve the diagnostic utility in this clinical setting compared with that of T1-weighted and STIR sequences alone for evaluation of the SI joint (21) and the spine (22). The importance of calibrating readers from different institutions prior to MRI reading exercises has been demonstrated in previous studies evaluating the SI joint and the spine in patients with SpA (23, 24).

Concordance among all 5 readers for the diagnosis of SpA by global assessment of the SI joint by MRI was 85% in the IBP group and 68% in the AS group. Similar concordance values for the 3 rheumatologists (89% and 78% for the 2 patient groups) compared with both radiologists (96% versus 87%) support the external validity of our study results. A study evaluating active and structural MRI changes in the SI joints of 68 patients with recent-onset IBP showed concordance rates between 2 readers of 78–85% for acute inflammation and 81–88% for structural changes (25).

In the AS group, erosion and fat infiltration in the SI joint as recorded by ≥2 readers were as frequent as bone marrow edema, while bone marrow edema was observed more frequently than structural lesions in the IBP group. However, erosions were detected in half of the IBP group with a mean symptom duration of 29 months; this finding indicates that structural damage of the SI joint starts early in the disease course, long before it can be captured by plain radiography. Also, fat infiltration was recorded in up to ∼44% of the IBP population; its specificity for SpA has been a matter of debate since the early reports on the use of MRI in the diagnostic evaluation of patients with SpA (26–29). Fat infiltration is not necessarily inflammation driven. Further work is needed to assess the diagnostic utility of fat infiltration per se or in combination with other MRI-visualized lesions suggestive for SpA.

Bone marrow edema and erosions as well as fat infiltration were recorded concordantly by ≥2 readers in up to 27% of control subjects with nonspecific back pain and 24% of healthy volunteers. This observation is consistent with a previous report describing the occurrence of bone marrow edema in the SI joints of 27% of patients with mechanical back pain and healthy control subjects (30) and 26% of healthy volunteers in a recent study assessing acute spinal lesions by MRI in early SpA (24). The pathophysiologic basis of these acute and structural MR changes remains speculative; the most likely explanations may be mechanically induced signal alterations or degenerative changes. In 2 persons with nonspecific back pain and 3 healthy control subjects, SpA was diagnosed by some readers according to the global assessment of the MRI scans. Ethical reasons prevented further examination of these 5 individuals, for instance, by testing for HLA–B27. The clinical implication from these observations is that low-grade acute lesions of the SI joint visible on MRI should be interpreted with caution to avoid misclassification of young persons with back pain as having SpA. Compared with bone marrow edema–like lesions and fat infiltration, erosions were observed much less frequently in both control groups, and they may contribute most to improved specificity.

A clinically relevant finding of our study was that the specificity obtained by global assessment of MRI of the SI joint, which reflects daily practice, was greater in the IBP group than that observed with the more detailed assessment of both the ASAS (10) and MORPHO study proposals. This may be partly explained by a possible reader bias when starting the detailed scoring after global assessment of the MRI: once the scan is regarded as negative for SpA, the reader may be less likely to record inflammatory lesions. However, both detailed evaluations showed a higher sensitivity, which is desirable in clinical practice to diagnose SpA in individual patients early in the disease course. The lower negative LR of the MORPHO study proposal compared with the other 2 approaches may prove useful in daily practice, where ruling out a diagnosis of SpA is often as important as confirming the disorder. Incorporating structural lesions in the detailed assessment defined by the MORPHO study proposal as opposed to the ASAS approach based only on bone marrow edema resulted in higher sensitivity while maintaining specificity. To our knowledge, the MORPHO study proposal represents the first data-driven definition of an MR image positive for SpA. Whether this evaluation system, encompassing both active and structural SI joint lesions, is able to enhance diagnostic utility must be confirmed in larger populations of patients with SpA, particularly in the group with preradiographic IBP.

Limitations of our study are the cross-sectional design and the relatively low number of patients with IBP, who represent the most challenging group of patients with SpA in daily practice, with the need to confirm the suspected diagnosis on clinical grounds. Long-term followup of patients with IBP will allow an estimation of the sensitivity and specificity of MRI according to the gold standard of radiographically defined sacroiliitis. The greater sensitivity and hence better diagnostic utility achieved by the radiologists in this study was mainly attributable to the more frequent recognition of structural lesions, suggesting that rheumatologists may require greater awareness and more intensive training to recognize structural lesions if they are to improve their ability to diagnose SpA on MRI.

In conclusion, this cross-sectional study, adopting a systematic and standardized evaluation of acute and structural lesions of the SI joint, showed that MRI has much greater diagnostic utility in SpA than has been documented previously. This study also showed that low-grade acute and structural SI joint lesions may be present in patients with nonspecific back pain and healthy volunteers, with a frequency of up to 27%. Future studies focusing on standardized evaluation of SI joint lesions should address the contribution of structural SI joint damage to diagnosis in early SpA. There is a need for greater awareness and training of rheumatologists in recognizing structural lesions on SI joint MRI.

AUTHOR CONTRIBUTIONS

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 published. Dr. Weber 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. Weber, Lambert, Hodler, Maksymowych.

Acquisition of data. Weber, Lambert, Østergaard, Hodler, Pedersen, Maksymowych.

Analysis and interpretation of data. Weber, Lambert, Østergaard, Hodler, Pedersen, Maksymowych.

Acknowledgements

We thank the patients and the healthy volunteers for their participation, and we thank the following Swiss rheumatologists, internists, and primary care physicians for referring their patients: A. Achermann (Luzern), M. Altermatt (Basel), C. Boetschi (Romanshorn), E. Bona (Chur), P. Bruehlmann (Zurich), C. Brunner (Zurich), A. Chamot (Morges), B. Elmiger (Bern), D. Galovic (Pfaeffikon), T. Gerber (Zurich), M. Giger (Menzingen), D. Glenz (Visp), F. Haefelin (Schlieren), G. Hajnos (Zurich), C. Harder (Luzern), U. Heusser (Winterthur), U. Hintermann (Brugg), M. Hoppler (Zug), P. Imbach (Zurich), J. Imholz (Zurich), C. Jeanneret (Schwerzenbach), D. Kaufmann (Zurich), B. Kleinert (Zurich), R. Kloeti (Brugg), I. Kramers (Zurich), R. Maager (Aarau), N. Masina (Lugano), C. Merlin (Baden), A. Rapp (Zurich), J. Ryser (Zurich), N. Satz (Zurich), A. Schmidt (Basel), H. Schwarz (Basel), S. Studer (Zurich), P. Sutter (Zurich), F. Tapernoux (Rueti), H. Trost (Zurich), B. Weiss (Basel).

We thank Tracey Clare, Clinical Research Manager, and Paul Filipow, Data Manager, Department of Radiology, University of Alberta, Edmonton, Canada for coordinating the Web-based SPARCC scoring index; Christian Streng, Balgrist University Hospital, Zurich, Switzerland for his technical assistance with Figure 1, and Rudolf O. Kissling, MD, Department of Rheumatology, Balgrist University Hospital, Zurich, Switzerland for scoring the sacroiliac joints (Balgrist patients) on plain radiographs.

Ancillary