Objectives To develop and test an MRI cartilage scoring system for use at the wrist in rheumatoid arthritis (RA).
Methods MRI scans were obtained using a 3T MRI scanner with dedicated wrist coil in 22 early and 16 established RA patients plus 22 controls. Axial and coronal T1-weighted (precontrast and postcontrast) and T2-weighted turbo spin echo sequences were obtained. Eight wrist joints were scored for cartilage narrowing: distal radioulnar, radiolunate, radioscaphoid, triquetrum-hamate, capitate-lunate, scaphotrapezoid, second metacarpal base-trapezoid and third metacarpal base-capitate, using a system based on the Sharp van der Heijde x-ray joint space narrowing (JSN) score by three radiologists. Fifteen sites at the wrist were also scored for synovitis, bone oedema and erosion using the RA MRI score.
Results Interobserver (three-reader) and intraobserver reliability (readers 1 and 2) for the cartilage score were excellent: intraclass correlations (ICC (95% CI)) 0.91, (0.86 to 0.94), 0.98 (0.96 to 1.00) and 0.94 (0.87 to 1.00), respectively. Cartilage scores (median, range) were higher in the established RA group (11.9, 2.3–27.3) than the early RA group (2.15, 0–6) (p≤0.001) but early RA scores did not differ from healthy controls (2.3, 1–8.7). Cartilage scores correlated with synovitis (R=0.52), bone oedema (R=0.63) and erosion scores (R=0.66), p<0.001 for all, and with x-ray JSN scores (R=0.68 to 0.78).
Conclusion This MRI cartilage score demonstrated excellent reliability when tested in a three-reader system. However, cartilage loss in early RA could not be distinguished from that seen in healthy controls.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
MRI is well established as an important imaging modality in the assessment of joint inflammation and damage in rheumatoid arthritis (RA). The rheumatoid arthritis MRI score (RAMRIS)1 has allowed the various pathologies within the rheumatoid joint to be quantified including synovitis, bone oedema and bone erosion. However, the assessment of cartilage change at the wrist using MRI in this disease has previously been hindered by relatively poor resolution at the cartilage/bone interface on 1.0T and 1.5T images. The OMERACT MRI working party initially included a cartilage score within the RAMRIS system but this was later deleted because of poor interreader reliability.2 Since that time, scanner technology has advanced, particularly with the advent of high field 3T MRI scanners with specialised surface coils that allow extremely high resolution and image clarity.3 With the addition of advanced imaging software and cartilage-specific sequences,4 feasibility issues may have been surmounted and it is now appropriate to revisit the use of MRI as a tool to quantify cartilage damage in RA. We have applied the Sharp van der Heijde (SvH) system that was developed for measuring joint space narrowing (JSN) on plain radiographs5 to 3T MRI imaging and have tested its reliability in three groups: early RA patients, established RA patients and healthy controls. We have also investigated correlations between cartilage change in RA and other components of the RAMRIS system including bone erosion, bone oedema and synovitis.
Patients and controls
Patients and controls were recruited with the approval of the New Zealand Multiregion Ethics Committee and all provided written informed consent. Those enrolled comprised 38 RA patients who fulfilled 1987 American College of Rheumatology criteria for RA6 and were either rheumatoid factor (RF) or anti-cyclic citrullinated peptide (anti-CCP) antibody positive, of whom 22 had early RA (duration <2 years) and 16 had established RA (duration 5–35 years). Patients were only enrolled following demonstration of normal creatinine clearance as administration of contrast was part of the MRI protocol. In addition, 22 age and sex-matched healthy volunteers were recruited as controls. Demographics including disease activity measures and medications for patients are presented in table 1.
MRI images were obtained on a 3T scanner (Philips Achieva 3T; Philips Electronics (Healthcare), The Netherlands). An eight-element SENSE phased array coil (receive only) was used. The dominant hand was placed in the wrist coil where it fitted snugly by the patient's side, palm facing the body, thumb anteriorly. The field of view was restricted to the carpus, including the distal radioulnar joint, extending to the metacarpal bases but excluding the metacarpophalangeal joints; dimensions 864×864 mm. MRI operational parameters are listed in table 2 and included a median slice thickness of 2.1 mm and matrix 400×307. Images were reconstructed in three dimensions for viewing at a matrix of 1024. The following turbo spin echo sequences were used: T1 weighted (T1w) and T2 weighted (T2w) sequences with fat saturation (FS) in the axial and coronal planes and proton density (PD) coronals (without FS) including an ultra high resolution sequence. T1wFS axial and coronal sequences were obtained postintravenous gadolinium diethylenetriamine pentaacetic acid given at a standard dosage of 10 ml Omniscan (Gadodiamide; 5.0 mmol/10 ml or 2.87 g/10 ml; GE Healthcare Inc, Princeton, New Jersey, USA). In controls, scans were performed without contrast for ethical reasons and no assessment of MRI synovitis was attempted.
Plain radiographs of the hands and feet were obtained in all patients but not normal controls for ethical reasons. Radiographs were scored by a rheumatologist (ND), blinded to clinical and MRI data, for erosions and JSN using the SvH score.5 Data were organised so that sum scores were obtained locally (dominant hand and wrist) and globally (hands and feet). Ten scans from the established RA group were re-read for intraobserver reliability.
Scoring MRI scans
Eight sites were chosen within the wrist where cartilage could be profiled well including: (1) distal radio-ulnar joint; (2) radiolunate joint; (3) radioscaphoid joint; (4) triquetrum-hamate joint; (5) capitate-lunate joint; (6) scaphotrapezoid joint; (7) second metacarpal base-trapezoid joint and (8) third metacarpal base-capitate joint (figure 1). Cartilage was scored in a blinded manner using a system based on the SvH score for radiographic JSN5 as follows: 0 (normal thickness); 1 (asymmetrical or minimal narrowing to maximum of 25%); 2 (definite narrowing with loss of up to 50% of the normal space); 3 (definite narrowing with loss of 50–99% of the normal space or subluxation) and 4 (absence of joint space, presumed ankylosis or complete luxation). Images illustrating each grade of cartilage change at the radiocarpal joint and are shown in figure 2. Scores were obtained independently by three radiologists expert in musculoskeletal imaging (AC, AM, QR). Score sheets included sample coronal MRI images (from a healthy control) indicating the sites and the optimal slice. Intraobserver reliability for cartilage scoring was obtained for readers 1 and 2. Scans were rescored 2 weeks later by reader 1 and 3 months later by reader 2, without access to previous scores. Images were also scored in a blinded manner for MRI erosion, bone oedema and synovitis (the latter in patients only) using the RAMRIS system.1 The EULAR–OMERACT RA MRI reference image atlas was used as an aid. Radiologists scoring MRI scans did not have access to x-ray images.
Intraclass correlations (ICC) with 95% CI were obtained for the three readers for MRI cartilage and other RAMRIS component scores for intraobserver and interobserver reliability. Calculations used a two-way random effects model assuming that the raters were a representative sample of a population of similar raters.7 A linear mixed model allowing for correlations between measurements by the same rater was used to investigate whether the cartilage scores differed between early and established RA groups. A square root transformation was used for cartilage scores to satisfy better the assumptions of normality. Spearman's correlations were used to investigate associations between the cartilage and RAMRIS synovitis/bone oedema/bone erosion scores. Spearman's correlations were used to investigate associations between SvH JSN scores and MRI cartilage scores.
MRI cartilage scores in RA patients and controls
Cartilage was most effectively imaged using PD sequences without FS as layers of cartilage overlying the carpal bones could be detected, separated by a black line when imaged perpendicular to the slice. Contrast-enhanced T1w images were useful to differentiate cartilage from enhancing synovium. Figure 3 shows examples of normal cartilage in a control compared with reduced cartilage thickness in RA patients. Cartilage scores averaged over the three raters (median, range) were higher in the established RA group (11.9, 2.3–27.3) than the early RA group (2.2, 0–6) (p<0.001) but early RA cartilage scores did not differ from healthy controls (2.3, 1–8.7) (figure 4).
Reliability measures for radiography and MRI cartilage scores
Intraobserver reliability for x-ray scoring (SvH score) was obtained by blinded rescoring of x-rays (by ND) from 15 patients from the early and established RA groups. Reliability was excellent for JSN (hands and feet): ICC 0.96 (95% CI 0.87 to 0.99), JSN (scanned hand), ICC 0.91 (0.76 to 0.97), erosion plus JSN score (scanned hand), ICC 0.93 (0.80 to 0.98). ICC for intraobserver reliability for MRI cartilage scores are shown in table 3 and were very high: reader 1: 0.98 (0.96 to 1.00) and reader 2: 0.94 (0.87 to 1.00). Intraobserver ICC for other aspects of the RAMRIS score were all greater than 0.8, (table 3). The three-reader interobserver ICC for cartilage scoring was also excellent: ICC (95% CI) 0.91 (0.86 to 0.94) (table 3). For other RAMRIS components, interobserver ICC were as follows: bone erosion 0.79 (0.61 to 0.88), bone oedema 0.80 (0.50 to 0.91) and synovitis 0.63 (0.46 to 0.77). Each reader pair were separately analysed and cartilage score ICC (95% CI) for each pair were similar to the three-way analysis as follows: reader 1 versus reader 2: 0.90 (0.85 to 0.95); reader 2 versus reader 3: 0.92 (0.88 to 0.96) and reader 1 versus reader 3: 0.91 (0.87 to 0.95). This also applied to other RAMRIS scores (data not shown).
MRI cartilage scores correlated with other RAMRIS measures
MRI cartilage scores for each reader correlated with their scores for RAMRIS synovitis (r=0.52), bone oedema (r=0.63) and bone erosion scores (r=0.66), p<0.001 for all. JSN scores from x-rays of the dominant hand (reflecting cartilage width) were also correlated with RAMRIS parameters (range for three readers) but to a lesser extent as follows: JSN (hand) versus RAMRIS synovitis (r=0.37–0.58), bone oedema (r=0.40–0.59) and bone erosion scores (r=0.53–0.59), p<0.02 for all.
MRI cartilage scores correlated with radiographic JSN scores
MRI cartilage scores (dominant wrist) were compared with local (dominant hand) and global (hands and feet) x-ray JSN scores. Results are presented in table 4. MRI cartilage scores were highly correlated with local (r=0.61–0.74) and global scores (r=0.68–0.78).
This is the first multireader study to show that the measurement of cartilage thickness using MRI scanning in RA wrists is feasible and reliable. Whereas scoring cartilage was attempted previously by the OMERACT MRI working party,2 reproducibility was poor, with an interobserver ICC of 0.18, leading to this component being discarded. On MRI, the bony cortex is represented as a ‘dark’, low signal line on T1w images8 adjacent to intermediate signal from cartilage, and this makes assessment of fine detail difficult, especially as the articular cartilage between carpal bones is thin. However, with advances in technology, including the improved signal-to-noise ratio of 3T scanners and appropriate surface coils, we have shown that the MRI assessment of cartilage damage at the wrist can be achieved with high interreader reliability, as indicated by the ICC of 0.91. In this study, although cartilage on both sides of the joint could be visualised (figure 3), the thickness of the two apposed cartilage layers was assessed as a single unit for practicality and feasibility. The sequences chosen did not allow changes in cartilage signal to be assessed.9 Many of the joints were the same as scored for JSN by the SvH system, but not all.5 This is because the three-dimensional MRI view differs from the two-dimensional radiographic view. We therefore did not attempt a joint-by-joint comparison between x-ray and MRI but compared totals for the wrist MRI cartilage score and the ‘scanned hand’ portion of the SvH JSN score as well as for hands and feet. We found strong correlations between MRI and x-ray methods.
There are few MRI studies of cartilage in RA. Gandy et al10 investigated cartilage volume at the knee using a 1.0 MRI scanner in 23 RA patients, and performed longitudinal imaging over 12 months in a subgroup. They used manual segmentation, delineating cartilage boundaries with automated volume estimation. This was highly reproducible, with only 5% variability between readers. Similarly, Eckstein et al3 investigated cartilage imaging using MRI volume segmentation at the knee in women with osteoarthritis, and concluded that the technology was sufficiently robust to be recommended for large-scale multicentre trials. Peterfy et al11 recently reported a study of JSN at the hand and wrist in 47 RA patients involved in two multicentre biologicals trials examining baseline MRI scans and radiographs. Using plain radiography as a gold standard, they found the sensitivity of MRI for JSN to be 0.94, specificity 0.91 and accuracy 0.91. As this was a single-reader study no assessment of reproducibility was possible.
The three readers in this study were radiologists, two of whom were working in the same institution. However, all read the scans separately and in a blinded manner following a training period. When data were examined between reader pairs, ICC were not appreciably different between readers from the same or different institutions, suggesting that reliability should be achievable by any trained reader using similar high quality MRI scans. The other components of the RAMRIS system were scored without any training, but all readers could refer to the EULAR–OMERACT MRI atlas.12 This resulted in high interobserver ICC for bone erosion and bone oedema but only moderate scores for synovitis. This may be because synovitis at the wrist can be difficult to score with high reproducibility without previous training, whereas reproducibility for synovitis scoring at metacarpophalangeal joints is generally higher.13
In view of the fact that many patients with RA have concomitant osteoarthritis, which might influence the cartilage score, we included 22 healthy controls who were matched as well as possible for age and sex and would be expected to have a similar incidence of background osteoarthritis. Interestingly, the median (range) cartilage score for controls (2.3, 1–8.7) was not significantly different from early RA patients (2.2, 0–6) with a disease duration of less than 2 years. As expected, the mean and range of MRI cartilage scores was higher in the established RA group (11.9, 2.3–27.3). Our data suggest that this lack of difference between the early RA and control groups could confound its use in the early RA setting, as the OMERACT filter requirement for sufficient between-group discrimination would not be satisfied.14 The mean age of patients in the established RA group was higher than controls, and as thinning of cartilage may occur with age, better matching of age would have been optimal. Clearly, longitudinal studies using larger numbers in patient and control groups would be required to investigate these issues.
Our findings raise questions regarding the timing of cartilage damage in RA, as traditionally this has been assumed to occur very early.15 MRI has provided new information about the early features of RA, and high scores for synovitis, tenosynovitis, bone oedema and erosions may occur in up to 45% of patients within 6 months of symptom onset.16 The relative lack of cartilage change in our early RA group, with a mean disease duration of 12 months, contrasts with their extensive synovial and bone changes, and suggests that erosion and/or thinning of articular cartilage may not be a feature of early disease at all. If the process of bone erosion depended on initial cartilage loss followed by infiltration by synovium and dissolution of bone matrix by degradative enzymes, then bone erosion should occur later than cartilage change and not vice versa. Our results are therefore more concordant with the hypothesis that bone erosion is driven by osteitis (bone oedema on MRI)17 and that cartilage change is a separate feature contributing to joint damage at a later date. Interestingly, Cohen et al18 recently reported a study investigating the efficacy of denusomab, which inhibits osteoclast-mediated bone resorption by targeting the receptor for activated nuclear factor κ B (RANKL). These authors found a reduction in the erosion component of the total Sharp score in RA patients treated over 12 months but no change in the JSN component, and concluded that bone erosion and cartilage damage may proceed by means separate pathways. A longitudinal follow-up of our patient cohort is planned to define the time-course of cartilage change and its relationship to the other features of rheumatoid pathology.
The authors wish to acknowledge the assistance of the staff of the Specialist Radiology and MRI Ltd, Greenlane, Auckland, for clerical and technical assistance and performing MRI scans. The expert knowledge and advice of Dr Ron Shnier (National Director of Research and Professional Development for Symbion Imaging) is also acknowledged. The authors also wish to thank the following rheumatologists for referring patients to this study: Dr Mike Butler, Dr Peter Gow, Dr Julia Martin, Dr Roger Reynolds, Dr Sunil Kumar, Dr Raoul Stuart, Dr Terry Macedo and the authors acknowledge the expert clinical assistance of nurse specialists Mrs Maria Lobo and Mrs Hazra Sahid.
Funding This work was supported by grants from the Auckland Medical Research Foundation, the Auckland Regional Rheumatology Research Trust and the University of Auckland (funded studentship for JD).
Competing interests None.
Patient consent Obtained.
Ethics approval This study was conducted with the approval of the New Zealand Multiregion Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.