Quantitative characterization of bone marrow edema pattern in rheumatoid arthritis using 3 tesla MRI




To develop imaging techniques that provide quantitative characterization of bone marrow edema pattern (BME) in wrist joints of patients with rheumatoid arthritis (RA), including volume, signal intensity changes, and perfusion properties.

Materials and Methods:

Fourteen RA patients and three controls were scanned using 3 Tesla MR. BME was semi-automatically segmented in water images obtained from iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL) sequences. BME perfusion parameters (enhancement and slope) were evaluated using three-dimensional (3D) dynamic enhanced MRI (DCE-MRI). Experimental reproducibility, inter- and intra-observer reproducibility of BME quantification were evaluated using root mean square coefficients of variation (RMS-CV) and intraclass correlation (ICC).


The RMS-CV for BME volume quantification with repeated scans were 6.9%. The inter-observer ICC was 0.993 and RMS CV was 5.2%. The intra-observer ICC was 0.998 and RMS CV was 2.3%. Both maximum enhancement and slope during DCE-MRI were significantly higher in BME than in normal bone marrow (P < 0.001). No significant correlation was found between BME quantification and clinical evaluations.


A highly reproducible semi-automatic method for quantifying BME lesion burden in RA was developed, which may enhance our capability of predicting disease progression and monitoring treatment response. J. Magn. Reson. Imaging 2012;35:211-217. © 2011 Wiley Periodicals, Inc.

RHEUMATOID ARTHRITIS (RA), a chronic inflammatory arthritis which affects an estimated 1.3 million Americans, is an important cause of long-term disability (1). RA causes synovial inflammation, damages articular cartilage, and erodes bone adjacent to joints. In the absence of effective therapy these structural abnormalities are progressive and lead to joint deformities and functional impairment over the time course of years. An important therapeutic goal in RA is to prevent joint deformities and subsequent disability.

During the past decade, there has been substantial progress in the treatment RA. Disease-modifying anti-rheumatic drugs (DMARDs) such as methotrexate (MTX) and biological agents, used individually or in combination, can slow and even halt bone erosion, but responses vary among individual patients. With current clinical assessments, it is difficult to determine whether a therapeutic regimen has halted joint damage in an individual patient; indeed, erosive disease can progress despite remission by standardized measures of disease activity (2, 3). Therefore, reliable identification of patients at greatest risk of disease progression and accurate monitoring and assessments of the response to these treatments have become increasingly important.

Radiography, the current “gold standard” for assessing structural joint damage in RA, does not visualize the earliest stages of erosive changes. MRI has been widely used in imaging RA (4, 5). It offers excellent contrast of synovial membrane, cartilage and bone. MRI provides the clinician with information about joint inflammation and damage and is more sensitive than conventional radiography. Furthermore, the semi-quantitative Outcome Measures in Rheumatology Clinical Trials (OMERACT) RA MRI scoring (RAMRIS) system has been established by an international collaborative effort for measurement of RA of the wrist and MCP joints (4).

Bone marrow edema (BME) patterns, the regions within trabecular bone with hyperintensity on T2-weighted fat-suppressed MRI, are commonly seen in RA. BME appears to correspond to areas on inflammation in subchondral bone and, possibly, with angiogenesis (6, 7). Multiple studies have suggested that BME is a strong predictor of progression of erosion and joint damage in RA (8–11). However, most studies on BME in RA have been limited to qualitative or semi-quantitative description of lesion size of BME, and few have investigated bone marrow perfusion (12). Quantitative characterization of BME would help to improve our ability to use these lesions as a prognostic imaging marker in RA.

Most previous MRI studies in RA used scanners primarily at 1.5 Tesla (T) or lower. Recent work using 3T MR shows improvement of image quality (13). In this study, we applied a 3T MR scanner with a dedicated wrist coil and advanced water fat separation sequences, which are critical to produce images with good quality for BME quantification. The goal of this study was to develop advanced imaging techniques that provide quantitative characterization of BME in wrist joints of patients with RA, including volume, signal intensity changes and perfusion properties, and to evaluate their reproducibility.



Fourteen RA patients (age: 53.7 ± 12.9 years with a range of 34–74 years and a median of 54 years, 11 females; disease duration: 93 ± 90.6 months with a range of 14–302 months and a median of 68 months) who fulfilled 1987 American College of Rheumatology (ACR) classification criteria for RA were recruited for the study. Thirteen patients were rheumatoid factor (RF) positive, and 13 patients had antibodies to cyclic citrullinated peptide (anti-CCP). Eight patients were on DMARD without biological treatment, while 6 were on combined DMARD and anti-tumor necrosis factor (anti-TNF) treatment. Standardized measurements of disease activity were collected during their clinical visit, including the 28 joint disease activity score (DAS28), and markers of inflammation (serum C-reactive protein [CRP] and erythrocyte sedimentation rate [ESR]). Three healthy subjects (age: 36.2 ± 17.2 years with a range of 25–56 years, 2 females) who had no history of diagnosed rheumatoid diseases were studied as controls. The study was approved by the Committee for Human Research at our institution. Written informed consent was obtained from all of the subjects after the nature of the examinations had been fully explained.

MR Imaging Protocols

All MR images were acquired at a 3T MR scanner (Signa HDx, GE Healthcare, Milwaukee, WI) with an eight-channel phased array wrist coil. Patients were positioned supine with arms resting on the side of the body. This position minimized potential motion and made it feasible to include patients with shoulder pain, which is difficult with the standard “superman” position. One challenge with this position is that the region of interest (ROI) will be significantly off-center, which may affect image quality, in particular fat suppression. We, therefore, compared fat-suppressed images acquired with conventional fast spin-echo (FSE) imaging to an advanced water/fat separation imaging technique using iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL) sequences (14). Briefly, IDEAL sequences, based on multipoint Dixon fat–water separation principle, uses an iterative linear least-squares method that decomposes water and fat images from source images acquired at short TE increments in one acquisition. IDEAL sequences provide images with high SNR and uniform water and fat separation. Further improvement in estimating fat more accurately, therefore better fat suppression in water images, was implemented by using multi-frequency fat spectrum reconstruction (15). Figure 1 shows representative images acquired in one RA subject with conventional FSE and IDEAL FSE sequences. The images illustrate that IDEAL FSE images provide superior fat suppression in bone marrow, which will help improve edema visualization and quantification. As an example, bone marrow edema in scaphoid of the RA subject in Figure 1 was quantified using the algorithm detailed below in the Image Processing section. The volume and signal intensity increase (relative to normal bone marrow) was 0.23 cm3 and 478.8% using IDEAL FSE images, and was 0.14 cm3 and 70.0% using conventional FSE images, respectively, indicating underestimate of edema using conventional FSE images.

Figure 1.

Fat-suppressed hand/wrist images of one RA subject at off-center location using conventional fast-spin echo (FSE) and IDEAL FSE images. a: T2-weighted fat-suppressed FSE images. b: T2-weighted FSE IDEAL water images with multi-frequency fat spectrum reconstruction, which shows superior fat suppression and was applied in this study.

Therefore, we applied IDEAL sequences with multi-frequency fat reconstruction in this study.

The imaging protocols included the major sequences recommended by Rheumatology Clinical Trials (OMERACT) as listed below:

  • Coronal and axial T2-weighted IDEAL FSE images (TR/TE = 3500/50ms, in plane resolution = 0.2 mm, slice thickness = 2 mm);

  • Coronal T1-weighted IDEAL FSE images (TR/TE = 600/9.9 ms, in plane resolution = 0.4 mm, slice thickness = 2 mm);

  • Coronal T1-weighted IDEAL spoiled gradient echo (SPGR) images (TR/TE = 15.3/2.9 ms, in plane resolution = 0.2 mm, slice thickness = 1 mm);

  • Coronal 3D dynamic contrast-enhanced (DCE) SPGR images acquired during Gd-DTPA injection (TR/TE = 6.4/2.1 ms, in plane resolution = 0.4 mm, slice thickness = 3 mm, temporal resolution 12 s, 32 time points, agent injection delay = 45 s);

  • Postcontrast coronal T1-weighted IDEAL FSE images (using the same parameters as precontrast T1-weighted IDEAL FSE images).

All images used GE ASSET parallel imaging with acceleration factor as two. The total scan time was approximately 45 min, and the total exam time including set up and positioning patients was less than one hour.

Image Processing

BME was identified independently by two radiologists in the T2-weighted IDEAL FSE water images and was segmented semi-automatically based on a thresholding algorithm developed previously (16) (Fig. 2a). First, contours (circles with 5 mm diameter) covering the normal bone marrow were placed manually, and the standard deviation (SD) of signal intensity within normal bone marrow was calculated. Second, a masked image was generated by manually drawing approximate contours of the bone containing BME. This procedure eliminated regions with high signal intensity outside bone marrow, such as soft tissues, pannus, and fluid. We excluded cysts, which were defined as well-defined round regions with very high signal intensity. Lastly, BME was automatically segmented with a threshold that was five times the SD of normal bone marrow using an in-house developed software based on IDL (Boulder, CO). BME volumes were calculated using 3D BME contours generated, and the signal intensity increase of BME was calculated as: (SIBME − SINBM)/SINBM × 100%, where SIBME is the mean signal intensity of BME and SINBM is the mean signal intensity of normal bone marrow. Perfusion parameters were calculated empirically based on the signal-time curve obtained from DCE-MRI (Fig. 2b): Maximum enhancement E = SImax/SIbase × 100% and Steepest Slope S = (SIafter−SIprior)/T × 100%, where SImax is the maximum signal intensity, SIbase is the baseline signal intensity before contrast agent is injected, SIprior is the signal intensity at the first time point, SIafter is the signal intensity at the second time point, and T is the time between SIprior and SIafter. A “baseline” image with signal intensity as an average of time point 0 to 5 during DCE-MRI sequence was registered to the T2w IDEAL FSE in-phase images (water + fat) rigidly using VTK CISG Registration Toolkit. The transformation matrix was then applied to the perfusion parameter map (slope and enhancement). BME ROIs were overlaid to the registered perfusion parameter maps. The average perfusion parameters were calculated within and outside BME for each subject.

Figure 2.

a: Semi-automatic segmentation of bone marrow edema (BME) pattern in wrist joints with RA. Five times standard deviation of signal within normal appearing bone marrow (blue ROI) were used to calculate threshold to segment BME (red) automatically. b: Perfusion parameters, including maximum enhancement and slope, were calculated empirically based on the signal-time curve obtained from dynamic contrast enhanced (DCE) MRI.

Two subjects were scanned twice for evaluating inter-experimental reproducibility. In four subjects (with seven BME lesions), the BME was segmented twice by one observer for evaluating intra-observer reproducibility; in eight subjects (with 12 BME lesions), the BME was segmented separately by two observers for evaluating inter-observer reproducibility. Both intra-observer and inter-observer reproducibility experiment involves separate placement of ROIs in normal bone marrow, and separate drawing of lines around BME lesions.

Statistical Analysis

The root mean square coefficients of variation (RMS-CV) of BME quantification were calculated for evaluating inter-experimental reproducibility:

equation image

Both inter- and intra-observer reproducibility were evaluated using intraclass correlation coefficients (ICC) and RMS-CV. BME volume and signal intensity were compared between carpal bones and distal radius/ulna using signed rank test (paired). Perfusion measurements (slope and maximum enhancement) within BME were compared with those outside BME using signed rank test (paired). Spearman correlation coefficients between imaging measures and clinical evaluation values (including DAS scores, ESR, and CRP) were calculated. The software R (www.r-project. org/) was used, and the significance level was 0.05.


The mean CRP was 4.8 ± 5.4 mg/l, the mean ESR was 24.8 ± 26.6 mm/h, and the mean DAS28 was 3.5 ± 1.6 for the RA patients.

Eleven of 14 RA patients showed BME. Of the patients who showed no BME, 2 were on DMARD treatment only and 1 was on DMARD and anti-TNF treatment. All of these 11 patients had BME in carpal bones, and 7 patients had additional BME in distal radius and/or distal ulna. No BME was observed in healthy controls.

The RMS-CV for BME volume quantification with repeated scans was 6.9%. The inter-observer ICC was 0.993 and RMS CV was 5.2%, the intra-observer ICC was 0.998 and RMS CV was 2.3%, respectively, indicating excellent reproducibility.

The mean total volume of BME was 0.790 ± 0.963 cm3. The mean signal intensity changes of BME compared with normal bone marrow was 635.8% ± 125.6%. In patients with BME both in carpal bones and distal radius/ulna, the total volume of BME within carpal bones were significantly higher than those in distal radius/ulna (P = 0.016; Table 1), while no significant difference was found in BME signal intensity change between these two locations (P > 0.05, Table 1).

Table 1. Volume and Signal Intensity Increase of BME Within Carpal Bones and Within Distal Radius/Ulna
 BME within carpal bonesBME within distal radius/ulnaP value
Volume (cm3)0.62 ± 0.790.26 ± 0.320.016
Signal intensity increase (%)630.0 ± 143.2644.9 ± 101.90.62
Maximum enhancement (% of baseline)144.9 ± 11.0128.3 ± 19.80.25
Slope (% of baseline/minute)188.1 ± 50.1102.5 ± 37.80.16

Both the maximum enhancement and slope of signal intensity curve during DCE MRI were significantly higher in BME than in normal bone marrow (140.4% ± 15.0% of baseline versus 111.1% ± 5.1% of baseline, P < 0.001 for maximum enhancement, and 164.7% ± 60.3% of baseline/minute versus 97.1 % ± 32.7% of baseline/minute, P < 0.001 for slope), Figure 3. The maximum enhancement and slope were slightly higher in BME within carpal bones compared with those within distal radius/ulna, but the difference was not significant (P > 0.05, Table 1).

Figure 3.

a: Signal intensity curve of DCE MRI within bone marrow edema (BME) and normal bone marrow (NBM) in an example patient. Signal was normalized to S0. b: Maximum enhancement and slope in BME are significantly higher than those in NBM (*P < 0.05).

There were no significant correlations between BME quantification and clinical evaluations (all correlation coefficients between BME quantification and clinical evaluations were smaller than 0.2 and all P values > 0.05). Indeed, discrepancies between BME and clinical measures of the disease were observed. In this cohort, six patients sustained clinical remission (defined as DAS28 < 2.6). Only one among these six patients showed no bone marrow edema. The other five patients had bone marrow edema with volume ranging from 0.140 cm3 to 2.282 cm3. Figure 4 shows representative patient images.

Figure 4.

T2-weighted FSE IDEAL water images. a: Water images of a healthy control. b: Postcontrast water images of a patient with recently diagnosed RA treated with MTX 7.5 mg weekly and prednisone 10 mg daily; DAS28-ESR was 4.74, which corresponds to moderate disease activity. c: Postcontrast water images of a RA patient treated with MTX 20 mg weekly and with DAS28 of 0.76, which corresponds to clinical remission. Both RA patients showed significant BME, bone erosion and enhanced synovium.


Despite substantial progress in the treatment of RA, a central issue remains: how to quickly and accurately determine whether or not a therapeutic regimen has arrested joint damage. Previous studies suggest that BME may be a promising predictor of radiographic progression by many (8–11). In a 2-year randomized controlled trial of early RA (CIMESTRA), BME measured in the nondominant wrist by RAMRIS was a stronger predictor of radiographic progression than other imaging measures, such as synovitis, and clinical evaluations (10). In their more recent 5-year report, baseline MRI-BME again, together with total Sharp score (TSS) and anti-CCP positivity, predicted radiographic progression at 5 years (11). In the present study, we focused on developing imaging and image processing methodologies for quantitative characterization of BME. Quantitative, compared with qualitative or semi-quantitative, characterization of such lesions should enhance our capability of stratifying patients for individualized treatment, predicting disease progression and monitoring treatment response at early time points.

Adequate fat suppression is critical in identifying and quantifying BME in bone marrow. Factors including field inhomogeneity and motion can cause failure of fat suppression using conventional fat saturation methods. In this study, we scanned the patient in a supine position, with hands resting on the side of the body. This method reduces potential motion artifact and will enable inclusion of patients with shoulder pain who cannot sustain the “superman” position. However, such positioning has the disadvantage of imaging the wrist and hand at the off center position (approximately 150 mm). In this situation, conventional fat-saturated FSE images failed to provide sufficient fat suppression as shown in Figure 1a. Instead, IDEAL sequences, especially with multi-frequency fat reconstruction (Fig. 1b), were robust to provide uniform water and fat suppression. In addition, IDEAL sequences can provide water, fat, in-phase (water + fat), out-of-phase (water − fat) and field map simultaneously, which provides multi images with different contrast between tissues and may facilitate further image analysis (such as registration between DCE-MRI and IDEAL FSE images).

Based on the IDEAL water images with superior fat suppression in bone marrow, we developed a semi-automatic method to segment BME. The results showed excellent intra-observer and inter-observer reproducibility. The intra-observer reproducibility in this study is comparable and slightly better than the BME subscore of the established OMERACT RAMRIS scoring system (ICC > 0.9) (17), while the inter-observer reproducibility is substantially better than the OMERACT RAMRIS edema scoring (ICC = 0.32 − 0.95) (5, 17). During semi-quantitative grading, for BME with size falling in the “gray area” between grades, grading variation can be easily introduced by readers. In this study, several factors potentially contributed to the improved reproducibility including improved image quality obtained with high magnetic field of 3T, dedicated eight-channel phased array wrist coil and the multi-frequency IDEAL sequence, as well as the quantitative evaluation which helps to reduce subjectivity. Furthermore, quantification of bone marrow edema can be more sensitive in detecting small changes longitudinally compared with semi-quantitative evaluation. In the present study, the ROI within normal bone marrow were defined manually, which may introduce variation of the threshold used for BME segmentation and subsequently for BME quantification. In particular, the use of a multi-channel phased array coil, as compared to volume coils, may lead to more signal variation. In this study, ROIs from multiple regions of normal bone marrow were used to minimize such variations. The reproducibility results from this preliminary study needs to be examined and confirmed with larger scale studies.

Increased signal enhancement was observed within BME compared with normal bone marrow in the present study, which is consistent with findings by Hodgson et al (12). Previous histopathological studies reported a cellular infiltrate comprising lymphocytes and osteoclasts, as well as increased number of osteoclasts and increased receptor activated nuclear factor κ ligand (RANKL) expression within BME, which could mediate the development of erosions from the marrow toward the joint surface (6, 7). Evidence from animal models also indicate that such an infiltrate corresponds with MRI BME, suggesting BME as a site for important pathology driving joint damage in RA (18). In addition, angiogenesis is known to accompany cellular proliferation in rheumatoid synovial membrane by means of mediators such as vascular endothelial growth factor and platelet derived growth factor. If the subchondral bone/bone marrow were proposed as another site of cellular proliferation in RA, angiogenesis may also present at this site. The increased enhancement during DCE-MRI within BME is indicative of potential angiogenesis and may due to increased vascularity and increased capillary permeability caused by pro-inflammatory cytokines (6, 19). Of interest, Hodgson et al also showed in their study (12) that the DCE-MRI enhancement was significantly reduced after anti-TNF treatment, although there was no significant changes in RAMRIS score of BME. These results suggested that BME perfusion parameters can be a more sensitive marker for treatment response than BME morphology (such as size).

The quantification of BME in this small cohort of patients showed a relatively large range. For example, in patients who showed BME, the volume ranged from 0.07 cm3 − 2.89 cm3, indicating the heterogeneous nature of the disease. Although the volume of BME was significantly larger in carpal bones compared with distal radius/ulna, no significant difference was found in signal intensity changes or perfusion parameters of BME between carpal bones and distal radius/ulna, indicating BME at these two sites have rather similar properties.

No significant correlation was found between BME quantification parameters and clinical evaluations, including CRP, ESR and DAS28. Rather, in this preliminary study, there were discrepancies between BME burden and these measures of disease activity, with BME present in the majority of patients in remission by DAS28. These results suggest that a subset of patients who are classified clinically in remission may sustain subclinical inflammation, which is consistent with several previous reports (2, 3). Such findings are clinically important because these patients, even in states of low disease activity, may benefit from additional therapy. These observations demonstrate the need for a more sensitive and objective evaluation using quantitative imaging techniques to identify inflammation and joint damage in RA.

There are several limitations of the present study. The subject cohort is small, especially the controls who were also younger than the RA subjects. The results of this study need to be evaluated in larger scale studies and compared with the established scoring system (RAMRIS for example). The quantitative characterization of BME also needs to be evaluated in longitudinal studies for their capability to predict disease and erosion progression. Future studies investigating bone marrow edema as a prognostic marker in RA need to stratify patients with anti-TNF treatment or not due to the influence of anti-TNF treatment to erosion progression in RA (20). The developed semi-automatic method involves operator interaction of defining normal bone marrow and outlining the approximate bone regions that contain BME. Fully automation of these two procedures using advanced image processing techniques are currently under development and evaluation.

In summary, we have developed a highly reproducible method for quantification of bone marrow edema pattern in RA using high field MRI and advanced imaging sequences. Quantification of bone marrow edema pattern may provide a powerful tool to assist physicians when making decisions regarding appropriate treatment strategies on the individual patient level.


The authors thank Dr. Sarah B. Gratton, Dr. Jonathan Graf, Dr. Thelma Munoz, Melissa Guan, and Gus Del Puerto for their help with recruiting patients, and Daniel Kuo and Fei Liang for helping with collecting MR data. This work was funded by a UCSF Academic Senate Individual Investigator Research Grant and Radiology Seed Grant.