Correlation between computer-aided dynamic gadolinium-enhanced ...

Rheumatology 2012;51:134–143 doi:10.1093/rheumatology/ker220 Advance Access publication 10 November 2011

RHEUMATOLOGY

Original article Correlation between computer-aided dynamic gadolinium-enhanced MRI assessment of inflammation and semi-quantitative synovitis and bone marrow oedema scores of the wrist in patients with rheumatoid arthritis—a cohort study Mikael Boesen1,2, Olga Kubassova3, Rasmus Bouert1, Mette B. Axelsen4, Mikkel Østergaard4, Marco A. Cimmino5, Bente Danneskiold-Samsøe2, Kim Hørslev-Petersen6 and Henning Bliddal2 Abstract

CLINICAL SCIENCE

Objective. To test the correlation between assessment of inflammation using dynamic contrast-enhanced MRI (DCE-MRI) analysed by a novel computer-aided approach and semi-quantitative scores of synovitis and bone marrow oedema (BME) using the OMERACT-RA MRI Scoring (RAMRIS) system, in the wrist of patients with RA. Methods. Fifty-four RA patients had conventional and DCE-MRI of a symptomatic wrist using a low-field 0.2T extremity scanner. RAMRIS synovitis and BME of the wrist joint were done. DCE-MRI data were analysed in three ways: (i) in all images (fully automated approach), (ii) within a large extended region of interest (ROI) placed around the wrist joint (semi-automated approach) and (iii) within a small ROI placed in the area with most visual enhancement (semi-automated approach). Time spent on each procedure was noted. Spearman’s rank correlation test was applied to assess the correlation between RAMRIS and the computer-generated dynamic parameters. Results. RAMRIS synovitis (range 2–9), BME (range 0–39) and the dynamic parameters reflecting the number of enhancing voxels were significantly correlated, especially when an extended ROI around the wrist was used ( = 0.74; P < 0.01 for synovitis and  = 0.82; P < 0.01 for BME). The observer spent on average 20 min (range 12–25 min) to perform RAMRIS, including acquisition of the results in the database, and 8 min (range 7–10 min) to perform all above-mentioned computer-aided analyses. Conclusion. Computer-aided analysis of DCE-MRI data correlated with RAMRIS synovitis and BME and was twice as fast to perform. This technique may be useful for quick semi-automated assessment of joint inflammation, but needs further validation. Key words: MRI, dynamic contrast-enhanced MRI (DCE-MRI), computer-aided analysis, RAMRIS, synovitis, bone marrow oedema, tenosynovitis, wrist, metacarpophalangeal joints

1

Department of Radiology, 2Parker Institute, Frederiksberg Hospital, Frederiksberrg, Denmark, 3Image Analysis Ltd, Leeds, UK, 4 Department of Rheumatology, Glostrup Hospital, Glostrup, Denmark, 5 Department of Rheumatology, University of Genoa, Genoa, Italy and 6 Department of Rheumatology, King Chr X Hospital, Graasten, Denmark. Submitted 18 October 2010; revised version accepted 23 May 2011. Correspondence to: Mikael Boesen, Department of Radiology, Frederiksberg Hospital, Nordrefasanvej 57, Frederiksberg Hospital, 2000 Frederiksberg, Denmark. E-mail: [email protected]

Introduction In RA, conventional post-contrast MRI of the wrist and MCP joints can be used to assess joint inflammation [1, 2]. The MRIs are usually analysed using the OMERACT RA MRI Scoring (RAMRIS) system [1], which is reproducible and sensitive to change [3]. Unfortunately, its use in clinical practice is limited by the long time

! The Author 2011. Published by Oxford University Press on behalf of the British Society for Rheumatology. All rights reserved. For Permissions, please email: [email protected]

Computer-aided analysis of dynamic gadolinium-enhanced MRI

needed to perform the complete evaluation, including synovitis, bone marrow oedema (BME) and erosions. The RAMRIS synovitis and BME scores are semiquantitative evaluations of the volume of enhancing synovium (synovitis) and the volume of bone oedema in the wrist and/or the MCP joints. The subjective evaluation of the images is, however, a source of bias adding to both intra- and inter-reader variability, especially in untrained users [3]. Compared with other imaging modalities such as X-ray and US used in RA patients, conventional MRI gives better evaluation of the synovitis and erosions in complex joints, such as the wrist, and MRI is the only tool to provide the possibility of studying BME, which is true inflammatory osteitis [4–7]. Recently, it was shown that imaging measures of inflammation detected with US Doppler in RA wrists had the highest predictive value of future erosive outcome in patients with low 28-joint DAS (DAS-28) [8]. This finding suggests that measures of perfusion characteristics, such as US Doppler and, potentially, dynamic contrast-enhanced MRI (DCE-MRI) [9–15], might reflect the degree of inflammation better than the volume of enhancing synovium seen in contrast-enhanced MRI images. DCE-MRI is a technique based on the sequential acquisition of rapid MRI sequences before and during the infusion of a contrast agent. It has previously been used to evaluate synovial inflammatory activity in patients with RA in the knees showing that the steepness of the dynamic curves correlates better with histological synovial vascularity and inflammatory cell infiltrate than measures of the corresponding post-contrast-enhancing synovial volumes [16–18]. The steepness of the dynamic curve in the synovium has also been shown to be very sensitive to change after IA steroid injections in knee joints with arthritis [13, 18, 19] and recently the same observation was published for BME in wrists of RA patients starting anti-TNF-a treatment [20]. Accordingly, observations made with DCE-MRI potentially allow detection of the early change in perfusion and inflammation upon treatment, which seems to occur before change in synovial volume and BME is seen in conventional MRIs [13, 21, 22]. DCE-MRI has been tested on high- and low-field dedicated extremity scanners and seems capable of discriminating patients with clinically active disease from those in remission in both knee and wrist joints [18, 21, 23, 24]. Although DCE-MRI is a very sensitive tool, its use has been restricted by the lack of appropriate methods for data analysis. Conventionally, DCE-MRI data are analysed using region of interest (ROI)-based techniques, where either the entire synovium is manually outlined in one slice or an observer places a small ROI in the most enhancing part of synovium. A fixed time interval of 55 s after contrast enhancement begins has been used to calculate the steepness of the enhancement curve [18, 23]. In addition, the reproducibility of the method over time in patients scanned 3 days apart seems to be poor, at least for the rate of early enhancement (REE) on low-field DCE-MRI

www.rheumatology.oxfordjournals.org

data [25] and it has been shown that the size and position of ROI have a great impact on diagnostic accuracy where an ROI displacement by only a few millimetres might result in up to 30% variation in the results [26]. In consequence, in the absence of a standardized procedure for placing ROIs, this method may generate variable and not always reliable and reproducible results. DCE-MRI data can also be corrupted by the patient’s hand movements, especially in RA patients with tender joints during prolonged examination procedures. The movements cause artefactual enhancement and are a source of large variation in the mean dynamic curves obtained by the ROI method [27]. The problems associated with the ROI-based methods may be solved by introducing a computer-aided analysis software, such as DYNAMIKA (www.imageanalysis.org. uk) [28, 29]. DYNAMIKA incorporates algorithms for motion reduction, which reduces blur effect in the images and significantly increases the signal-to-noise ratio (SNR) [30]. This is followed by automated analysis of the data using a voxel-by-voxel approach, which can guide the ROI placement or can be used for fully automated data analysis [31]. This computer-aided method has proved very robust with high inter/intra-reader reproducibility [32]. The aim of this study was to compare the computer-aided inflammatory measures generated from DCE-MRI data on a low-field dedicated extremity scanner with the RAMRIS of synovitis and BME.

Material and methods Patients A cohort of 54 consecutive patients with RA, fulfilling the ACR 1987 criteria [33] were recruited from the outpatient clinic at the Department of Rheumatology, Frederiksberg Hospital. They had clinical involvement of at least one wrist and exclusion criteria were known contraindications for contrast-enhanced MRI, such as pacemakers, impaired renal function based on an estimated glomerular infiltration rate, etc. Age: median 52 years (range 24–79 years), disease duration: median 11 years (range 3–28 years) and DAS-28: median 4.0 (range 1.3–7.4). The patients were all on DMARDs and were seen in the outpatient clinic either as part of the annual control (n = 30), as part of the evaluation before potential start of biologic treatment (n = 14) or before a supplementary injection with IA steroid because of a flair in the wrist (n = 10). No patients had received an IA injection of steroid 3 months before the study. All patients included had MRI of the most symptomatic wrist including a dynamic contrast-enhanced sequence using a low-field 0.2T dedicated extremity scanner (Esaote E-scan, Genova, Italy). The study was approved by the local ethics committee (Committee of the Capital Region of Denmark, approval number KF 01-045/03) and was carried out in accordance with the Helsinki Declaration II and the European Guidelines for Good Clinical Practice.

135

Mikael Boesen et al.

MRI Static and dynamic MRI using a 0.2T musculoskeletal extremity E-scanner (ESAOTE Ltd) was performed in all 54 subjects. The patients were examined in supine position with the hand along the side of the body positioned centrally in the receive-only cylindrical solenoid wrist coil. To avoid large-movement artefacts, the patients were positioned as comfortably as possible with a supporting pillow under the forearm and elbow; the hand was fixed in the wrist coil with small supporting foam pillows. The imaging protocol was the following: gradient echo scout, coronal short tau inversion recovery (STIR) (TR/TE/TI: 1310/24/85, fov/matrix: 170  170 mm/192  128 and slice thickness 3.0 mm) and a coronal turbo 3D T1weighted (T1-w) gradient echo (TR/TE: 38/16, FOV/ matrix: 170  170  45 mm/192  160  52 and slice thickness 0.9 mm) all images before contrast. Starting exactly at the time of the i.v. injection of 0.1 mmol/kg body weight gadolinium (Gd) contrast (Multihance, Bracco S.p.A., Italy), 30 consecutive acquisitions of three pre-positioned 4-mm coronal T1-w gradient echo dynamic magnetic resonance images (TR/TE 60/6, FOV/imaging matrix 180  180 mm/256  256, FA 75, NEX 1) were performed. The three axial pre-positioned 4-mm DCE-MRI slices with no inter-slice gap were positioned tangentially to the long axis of the radius covering the central 1.2 cm of the wrist in the coronal plane. Each slice was obtained every 10 s, for 300 s. Finally, the coronal 3D T1-w gradient echo sequences were repeated. The acquisition time of each sequence ranged from 4 to 8 min, with one signal acquired. Total imaging time was 35 min. Before the images were sent to the picture archiving and communication system (PACS), axial pre- and post-contrast T1 gradient echo images of 0.9-mm slice thickness were reconstructed from the isotropic coronal slices using the scanner software.

Data analysis The STIR and pre-/post-contrast T1-w gradient echo images were used for RAMRIS scoring of BME and synovitis, respectively. DYNAMIKA software version 2.1.0.2 (www.imageanalysis.org.uk, Leeds, UK) was used to analyse the DCE-MRI data. Using this software, we reduced patient motion artefacts between the dynamic frames, which allowed reduction in artefactual enhancement, thus increasing the SNR by a factor of 2 (data not shown). Further, the data were analysed using the voxel-by-voxel-based approach incorporated into the software [31]. The enhancement pattern of each signal intensity vs time curve was recognized and further colour coded in the maps of Gd uptake, as one of four models: (i) no enhancement (no colour); (ii) persistent enhancement (blue); (iii) plateau, e.g. baseline followed by an uptake and then plateau (green); (iv) wash-out, e.g. tissues where signal intensity vs time curves exhibited a prominent wash-out phase. Each voxel signal intensity vs time curve was approximated by one of the models and the corresponding colours were superimposed on the greyscale pre-contrast

136

dynamic T1–w image (Figs 1 and 2) [13]. Additional parameters such as initial rate of enhancement (IRE), maximum enhancement (ME) and time of onset of enhancement (Tonset) were extracted from the automatically chosen appropriate model. The parametric maps of ME, IRE, Gd uptake pattern and Tonset are shown in Figs 1 and 2.

Understanding the dynamic enhancement maps The ME map (Fig. 1A) shows the intensity increase over a baseline in a particular voxel and ME is measured as a ratio between the baseline and the ME within the enhancement model. The IRE map (Fig. 1B) shows the increase in voxel intensity per second from Tonset until ME is reached. The highest numbers for ME and IRE are shown in bright yellow—white and lower values in redder colours. The colours in the Gd map (Fig. 1C) reflect the behaviour of the Gd over time as stated above, where voxels with no Gd uptake have no colour; voxels with continuous pattern of enhancement are shown in blue; voxels with plateau pattern of enhancement in green and voxels with wash-out pattern in red. The Gd map allows measurement of the number of pixels with a particular pattern of enhancement and these parameters and their combinations were used to calculate the total number or the volume of enhancing voxels (N-total), number of voxels reaching plateau enhancement (N-plateau) and wash-out enhancement (N-wash-out) in both the entire image and the drawn ROIs (see below). The Tonset map (Fig. 1D) shows the time in seconds where the enhancement curve begins compared with the first baseline frame in each voxel, where redder colours indicate the shortest time of onset. The vertical colour bars or the Y-axis in the four enhancement maps display the values of the chosen parameter (ME, IRE, Gd and Tonset; Figs 1 and 2). The values are measured in each voxel and then grouped into 10 equally spaced bins. These are displayed on the colour bar. The horizontal colour bar shows the number of voxels and their percentage of the total in parentheses for each statistic in the corresponding enhancement MAPs, e.g. 1604 (94.6%), 8 (0.5%), 68 (4%), 15 (0.9%) and 1 (0.1%) in Fig. 1A, etc. The white cross seen in the software screenshots of Figs 1 and 2 corresponds to a pointer in the software, where an enhancement curve and the associated enhancement statistics are generated from the voxel where it is placed. For more information visit www.imageanalysis.org.uk.

Analysing the DCE-MRI data To quantify the degree of inflammation in the wrists, the dynamic enhancement was analysed in three ways: (i) in all images (fully automated approach), (ii) within a large extended ROI to cover the wrist joint in the slice with most visual enhancement avoiding the larger blood vessels (Fig. 1) (semi-automated approach) using the maps of ME, IRE and Gd as guidance and (iii) within a small ROI

www.rheumatology.oxfordjournals.org

Computer-aided analysis of dynamic gadolinium-enhanced MRI

FIG. 1 Parametric maps of DCE-MRI data and ROI placement. Parametric maps of ME (A), IRE (B), Gd uptake (C) and Tonset (D) superimposed on a coronal T1-w dynamic image of the wrist derived from a patient with active RA. The ME, IRE and Gd maps are used to guide the ROI placement (encircled in blue). Larger blood vessels (white arrows) are excluded from the analysis using the extended wrist ROI. Vessels ( ) are tubular structures with wash-out pattern (red colour) on the Gd map (C) and bright high values of ME (A) and IRE (B). The white cross seen in the software screenshots corresponds to a pointer, where an enhancement curve and the associated enhancement statistics are generated from the voxel where it is placed.

placed in the area of most visual enhancement (semi-automated approach) in the same slice as (ii), again guided by the maps of ME, IRE and Gd. RAMRIS was done by an experienced musculoskeletal radiologist (M.B.) blinded to clinical and the dynamic imaging data. The dynamic data analysis using DYNAMIKA was also done blindly by an experienced user (R.B.). The correlation of all DYNAMIKA-derived parameters with the RAMRIS of synovitis and oedema were done using Spearman’s rank coefficient using SPSS statistics, version 19 (IBM).

Results Spearman’s rank correlation coefficient  between RAMRIS synovitis and BME and the DYNAMIKA-retrieved parameters are listed in Table 1. Comparing the DYNAMIKA parameters with the sum of RAMRIS synovitis and BME revealed similar correlation statistics to those of the correlation found for the individual RAMRIS parameters displayed in Table 1 (data not

www.rheumatology.oxfordjournals.org

shown). We found no statistically significant correlation between DAS-28 and either the RAMRIS or the DYNAMIKA parameters (data not shown). RAMRIS synovitis and BME were significantly correlated ( = 0.76, 95% CI 0.62, 0.86; P < 0.01). The observer spent on average 20 min (range 12–25 min) to score synovitis and BME in one data set using the RAMRIS method, divided between performing the score (8–10 min) and acquisition of the results in the database. Analysing the dynamic data using DYNAMIKA, the observer spent on average 8 min (range 4–10 min) to perform intra-study motion correction (2–4 min) and further analysis of the data using both the fully automated, the extended ROI and the small ROI analysis incorporated into the software.

Discussion In this study, analysing a group of RA patients with heterogeneous degrees of disease activity and inflammation, we found a moderate to high correlation between the

137

Mikael Boesen et al.

FIG. 2 MCP synovitis. Parametric maps obtained from a patient with low RAMRIS of synovitis in the wrist, but a high inflammatory activity in the second to fifth MCP joints. This was a confounder to correlation analysis between RAMRIS from the wrist and the fully automatic analysis of the dynamic images, which will include the enhancement seen in the MCP joints. The contribution from the MCP joints and tubular vessels (arrow) can easily be excluded using an extended wrist ROI (encircled in blue).

DCE-MRI-derived inflammation scores assessed by DYNAMIKA, and the semi-quantitative RAMRIS of synovitis and BME. The best correlations for RAMRIS synovitis were seen with the dynamic parameters describing the volume of enhancement [N-total, N-plateau and N-wash-out (Table 1)]—especially when an extended wrist ROI was applied in the slice with most visual activity, which in a recent study by our group showed high intraand inter-reader reliability [25]. The highest correlation was seen with the sum of N-plateau and N-wash-out ( = 0.74; P < 0.01) (Table 1). This is in accordance with the fact that RAMRIS synovitis is a semi-quantitative assessment of the enhancing synovial volume. For RAMRIS BME, we also found a moderate to high correlation with the dynamic data describing the volume of enhancement, again with the highest correlation seen using the sum of N-plateau and N-wash-out ( = 0.82; P < 0.01) within the extended wrist ROI. The observed correlation may partly be explained by the high correlation between the RAMRIS synovitis and BME scores ( = 0.76;

138

P < 0.01). However, another contributing reason could be that bone affected by BME enhances after Gd infusion in DCE-MRI data [20], and thus will be depicted by the software as voxels with contrast enhancement (Figs 3 and 4). For the dynamic parameters of ME and IRE, we only found a poor to moderate correlation with the RAMRIS of synovitis and BME, and with a wide CI. The highest correlation was seen when using a small user-defined ROI (Table 1), whereas there was only a poor correlation with the data derived from the fully automatic whole-joint assessment. The main reason for this is probably that vessels show fast and high Gd-contrast uptake in the IRE and ME maps with a wash-out phase appearing prominently as red pixels in the Gd maps (arrow in Figs 1 and 2). When the data are processed in a fully automated manner and the vessels are not excluded from the analysis, the statistics acquired from the entire joint will be skewed especially in the case of low disease activity. This means that the mean value of ME, IRE and the number of pixels with wash-out pattern (N-wash-out)

www.rheumatology.oxfordjournals.org

Computer-aided analysis of dynamic gadolinium-enhanced MRI

TABLE 1 Correlation between DCE-MRI analysis and RAMRIS synovitis and BME MRI analysis methods Fully automatic whole joint assessment IRE ME N-total N-plateau N-wash out N-plateau + N-wash out User-defined extended wrist ROI IRE ME N-total N-plateau N-wash out N-plateau + N-wash out User-defined small ROI IRE ME N-total N-plateau N-wash out N-plateau + N-wash out

RAMRIS synovitis, q-value (95% CI)

RAMRIS BME, q-value (95% CI)

0.23 0.24 0.61 0.59 0.55 0.72

(0.19, 0.47)* (0.15, 0.45)* (0.4, 0.76)** (0.4, 0.74)** (0.35, 0.72)** (0.61, 0.89)**

0.29 0.18 0.51 0.41 0.59 0.68

(0.21, 0.42)* (0.15, 0.51)* (0.4, 0.69)** (0.36, 0.61)** (0.4, 0.74)** (0.61, 0.91)**

0.39 0.47 0.68 0.71 0.66 0.74

(0.31, (0.31, (0.61, (0.64, (0.58, (0.61,

0.42)** 0.63) ** 0.73)** 0.78)** 0.71)** 0.85)**

0.37 0.33 0.65 0.66 0.80 0.82

(0.29, (0.28, (0.55, (0.58, (0.68, (0.62,

0.53) ** 0.35)** 0.72)** 0.81)** 0.81)** 0.84)**

0.41 0.58 0.33 0.16 0.12 0.24

(0.38, (0.41, (0.24, (0.12, (0.12, (0.11,

0.74)** 0.82)** 0.41)** 0.31)* 0.3)* 0.33)*

0.52 0.38 0.18 0.16 0.14 0.17

(0.43, (0.31, (0.11, (0.13, (0.11, (0.12,

0.72)** 0.52)** 0.36)* 0.31)* 0.32)* 0.34)*

*P < 0.05; **P < 0.01.

FIG. 3 Severe wrist synovitis and BME. Patient with severe RAMRIS synovitis (Score 9) and BME (Score 39). (A) STIR, (B) pre-contrast T1-turbo gradient echo, (C) post-contrast T1-turbo gradient echo and (D–F) corresponding parametric map from the same slice of ME (D), IRE (E) and Gd (F) enhancement pattern. Note the severe global BME on the STIR sequence, which is enhanced on the post-contrast T1-turbo image (C). The bone marrow enhancement is also depicted in the parametric maps generated from the DCE-MRI images. This could be an explanation for the high correlation between BME and the number of enhancing voxels.

www.rheumatology.oxfordjournals.org

139

Mikael Boesen et al.

FIG. 4 Moderate wrist synovitis and BME. Patient with moderate RAMRIS synovitis (Score 5) and BME (Score 9): (A) STIR, (B) pre-contrast T1-turbo gradient echo and (C) contrast-enhanced T1-turbo gradient echo. (D–F) Corresponding parametric maps from the same slice of ME (D), IRE (E) and Gd (F) enhancement pattern, respectively. Note the enhancing synovium in the wrist on the contrast-enhanced image in (C), as well as the BME in the scaphoid bone on the STIR sequence in (A). Note also the normal bone marrow signal in the majority of carpal bones. The parametric maps (D–F) also depict the synovial enhancement as well as the BME in the scaphoid, whereas there is no dynamic enhancement in the carpal bones with normal marrow signal. This may explain the high correlation between BME scores and the number of enhancing voxels.

will increase due to the enhancement characteristics of the vessels. In order to estimate the enhancement of the synovial tissue more objectively, these highly perfused vessels should be excluded from the evaluation. This could easily be done by drawing a quick extended wrist ROI, outlining the inflamed area and avoiding vessels (Fig. 1), which also leads to a higher correlation (Table 1). Using the fully automatic analysis, another possible confounder to the correlation was the coronal image orientation of the DCE-MRI data with a field of view of 140 mm, where often both the wrist and the MCP joints are visualized within the same image. The coronal image orientation allows good visualization of inflammatory disease severity in both joint areas, but when inflammation is present in the MCP joints the fully automatic dynamic analysis will overestimate the amount of inflammation compared with the RAMRIS scores in the wrist. This is especially the case when there is low synovitis and BME activity in the wrist, as illustrated in Fig. 2, showing dynamic maps from a patient with low RAMRIS synovitis and BME scores in the wrist but high activity in four ulnar MCP joints. The inflammatory contribution from the MCP joints could easily be excluded from the analysis by drawing a

140

quick extended wrist ROI, which in our study gave a higher correlation with the RAMRIS synovitis and in particular the BME scores (Table 1). A third confounder to the correlation analysis was tenosynovitis encountered in some of the included patients. Figure 5 provides an example of a patient with both moderate wrist synovitis and severe tenosynovitis. In this case, the software overestimates disease activity, especially the number of enhancing voxels, compared with RAMRIS of the wrist. However, including tenosynovitis in the analysis might be a potential advantage as this feature is part of the inflammatory load of the individual patient. This is supported in a recent publication showing that the combination of RAMRIS synovitis, BME and a novel tenosynovitis score revealed higher standardized response mean during anti-TNF-a treatment than each parameter alone [34]. This indicates that the enhancement pattern and disease activity in the wrist, MCP and tendons all contribute to the disease burden; however, often in clinical trials the focus is on one joint region without inclusion of tenosynovitis. If separation between synovitis and tenosynovitis is needed an axial image orientation covering the wrist only is recommended [23]. Future studies should aim at investigating the impact of inflammation on all relevant tissues, which might be

www.rheumatology.oxfordjournals.org

Computer-aided analysis of dynamic gadolinium-enhanced MRI

FIG. 5 Wrist synovitis and tenosynovitis. Patient with both wrist synovitis ( ) and tenosynovitis ( ). In this patient, the DYNAMIKA analysis will overestimate the disease activity compared with the RAMRIS, which does not include tenosynovitis.

estimated automatically using DCE-MRI and appropriate computer-aided software such as DYNAMIKA. This could lead to a better understanding and classification of both the disease burden and the individual treatment response in RA patients. In addition, our results need to be verified using both low-field 1.5T and 3.0T scanners. Using high-field scanners might give even better correlations with the RAMRIS as was shown for RAMRIS synovitis in a recently published study using 3T DCE-MRI of wrists in patients with undifferentiated polyarthritis. This finding could be due to higher SNR, which results in the ability to cover larger areas than the three slices limited to low-field scanners. A higher SNR could also be used to decrease the time between the dynamic frames, thus allowing a more precise curve based on a higher number of measuring points. Finally, the use of computer-aided software was approximately twice as fast to perform compared with the time used on average to score and register the RAMRIS data. This may lead to increased productivity in data analysis. In conclusion, results from computeraided analysis of DCE-MRI data correlated with the semi-quantitative RAMRIS of synovitis and BME and was twice as fast to perform. This technique may be useful for quick semi-automated assessment of joint inflammation, but needs further validation. Rheumatology key messages Computer-aided analysis of DCE-MRI data correlated with RAMRIS synovitis and BME was quick to perform. . The volume of enhancing voxels gave the best correlations with RAMRIS. . Computer-aided analysis of DCE-MRI also depicts tenosynovitis and MCP joint synovitis. .

Acknowledgements We would like to thank the OAK Foundation and the Danish Rheumatism Association for their financial

www.rheumatology.oxfordjournals.org

support to do this study, and we would like to thank Abbott A/S Denmark for an unrestricted grant to perform the analysis. Funding: OAK Foundation, Danish Rheumatism Association. Abbott A/S Denmark supported the study with an unrestricted grant, which was used to pay running and publication costs of the study including purchase of a DYNAMIKA licence. Disclosure statement: O.K. is a founder and managing director of Image Analysis Ltd, which developed and supported DYNAMIKA software used for dynamic MRI data analysis. Image Analysis Ltd team had no influence on the results published in this manuscript and only provided technical support with the use of the software DYNAMIKA. M.B. is a consultant for Image Analysis Ltd, Leeds, UK. All other authors have declared no conflicts of interest.

References 1 Ostergaard M, Peterfy C, Conaghan P et al. OMERACT rheumatoid arthritis magnetic resonance imaging studies. Core set of MRI acquisitions, joint pathology definitions, and the OMERACT RA-MRI scoring system. J Rheumatol 2003;30:1385–6. 2 Ostergaard M, Conaghan PG, O’Connor P et al. Reducing invasiveness, duration, and cost of magnetic resonance imaging in rheumatoid arthritis by omitting intravenous contrast injection – does it change the assessment of inflammatory and destructive joint changes by the OMERACT RAMRIS? J Rheumatol 2009;36:1806–10. 3 Haavardsholm EA, Ostergaard M, Ejbjerg BJ et al. Reliability and sensitivity to change of the OMERACT rheumatoid arthritis magnetic resonance imaging score in a multireader, longitudinal setting. Arthritis Rheum 2005; 52:3860–7. 4 Dohn UM, Ejbjerg BJ, Hasselquist M et al. Detection of bone erosions in rheumatoid arthritis wrist joints with magnetic resonance imaging, computed tomography and radiography. Arthritis Res Ther 2008;10:R25.

141

Mikael Boesen et al.

5 Conaghan PG, McGonagle D, Wakefield R et al. New approaches to imaging of early rheumatoid arthritis. Clin Exp Rheumatol 1999;17:S37–42. 6 Boesen M, Ellegaard K, Boesen L et al. Ultrasound Doppler score correlates with OMERACT RAMRIS bone marrow oedema and synovitis score in the wrist joint of patients with rheumatoid arthritis. Ultraschall Med 2011, Jan 21 [Epub ahead of print]. 7 McQueen FM, Ostendorf B. What is MRI bone oedema in rheumatoid arthritis and why does it matter? Arthritis Res Ther 2006;8:222. 8 Brown AK, Conaghan PG, Karim Z et al. An explanation for the apparent dissociation between clinical remission and continued structural deterioration in rheumatoid arthritis. Arthritis Rheum 2008;58:2958–67. 9 Boesen M, Ostergaard M, Cimmino MA et al. MRI quantification of rheumatoid arthritis: current knowledge and future perspectives. Eur J Radiol 2009;71:189–96. 10 Hodgson RJ, O’Connor P, Moots R. MRI of rheumatoid arthritis image quantitation for the assessment of disease activity, progression and response to therapy. Rheumatology 2008;47:13–21. 11 Huang J, Stewart N, Crabbe J et al. A 1-year follow-up study of dynamic magnetic resonance imaging in early rheumatoid arthritis reveals synovitis to be increased in shared epitope-positive patients and predictive of erosions at 1 year. Rheumatology 2000;39:407–16. 12 Malattia C, Damasio MB, Basso C et al. Dynamic contrast-enhanced magnetic resonance imaging in the assessment of disease activity in patients with juvenile idiopathic arthritis. Rheumatology 2010;49:178–85. 13 Ostergaard M, Stoltenberg M, Henriksen O et al. Quantitative assessment of synovial inflammation by dynamic gadolinium-enhanced magnetic resonance imaging. A study of the effect of intra-articular methylprednisolone on the rate of early synovial enhancement. Br J Rheumatol 1996;35:50–9. 14 Palosaari K, Vuotila J, Takalo R et al. Contrast-enhanced dynamic and static MRI correlates with quantitative 99Tcm-labelled nanocolloid scintigraphy. Study of early rheumatoid arthritis patients. Rheumatology 2004;43: 1364–73. 15 Palosaari K, Vuotila J, Takalo R et al. Bone oedema predicts erosive progression on wrist MRI in early RA–a 2-yr observational MRI and NC scintigraphy study. Rheumatology 2006;45:1542–8. 16 Gaffney K, Cookson J, Blake D et al. Quantification of rheumatoid synovitis by magnetic resonance imaging. Arthritis Rheum 1995;38:1610–7. 17 Ostergaard M, Stoltenberg M, Lovgreen-Nielsen P et al. Magnetic resonance imaging-determined synovial membrane and joint effusion volumes in rheumatoid arthritis and osteoarthritis: comparison with the macroscopic and microscopic appearance of the synovium. Arthritis Rheum 1997;40:1856–67. 18 Ostergaard M, Stoltenberg M, Lovgreen-Nielsen P et al. Quantification of synovitis by MRI: correlation between dynamic and static gadolinium-enhanced magnetic resonance imaging and microscopic and macroscopic signs of synovial inflammation. Magn Reson Imaging 1998;16: 743–54. 19 Boesen M, Kubassova O, Cimmino MA et al. Dynamic contrast enhanced MRI can monitor the very early

142

inflammatory treatment response upon intra-articular steroid injection in the knee joint: a case report with review of the literature. Arthritis 2011:1–8, doi:10.1155/2011/ 578252. 20 Hodgson R, Grainger A, O’Connor P et al. Dynamic contrast enhanced MRI of bone marrow oedema in rheumatoid arthritis. Ann Rheum Dis 2008;67:270–2. 21 Ostergaard M, Lorenzen I, Henriksen O. Dynamic gadolinium-enhanced MR imaging in active and inactive immunoinflammatory gonarthritis. Acta Radiol 1994;35: 275–81. 22 Ostergaard M, Stoltenberg M, Gideon P et al. Changes in synovial membrane and joint effusion volumes after intraarticular methylprednisolone. Quantitative assessment of inflammatory and destructive changes in arthritis by MRI. J Rheumatol 1996;23:1151–61. 23 Cimmino MA, Innocenti S, Livrone F et al. Dynamic gadolinium-enhanced magnetic resonance imaging of the wrist in patients with rheumatoid arthritis can discriminate active from inactive disease. Arthritis Rheum 2003;48:1207–13. 24 Navalho M, Resende C, Rodrigues AM et al. Dynamic contrast-enhanced 3-T magnetic resonance imaging: a method for quantifying disease activity in early polyarthritis. Skeletal Radiol 2011, Feb 12 [Epub ahead of print]. 25 Boesen M, Kubassova O, Parodi M et al. Comparison of the manual and computer-aided techniques for evaluation of wrist synovitis using dynamic contrastenhanced MRI on a dedicated scanner. Eur J Radiol 2011; 77:202–6. 26 McQueen FM, Crabbe J, Stewart N. Dynamic gadoliniumenhanced magnetic resonance imaging of the wrist in patients with rheumatoid arthritis: comment on the article by Cimmino et al. Arthritis Rheum 2004;50:674–5. 27 Kubassova O, Boyle RD, Boesen M. Impact of registration on enhancement curve estimation in DCE-MRI data. In Proceedings Workshop on Functional Image Analysis of International Conference on Medical Image Computing and Computer Assisted Intervention, 9-1-2008. pp. 57–65, http://eprints.whiterose.ac.uk/5394/ (11 July 2011, date last accessed). 28 Kubassova O, Boesen M, Boyle RD et al. Fast and robust analysis of dynamic contrast enhanced MRI datasets. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv 2007;10: 261–9. 29 Kubassova OA, Boyle RD, Radjenovic A. Quantitative analysis of dynamic contrast-enhanced MRI datasets of the metacarpophalangeal joints. Acad Radiol 2007;14: 1189–200. 30 Boesen M, Kubassova O, Bliddal H et al. Automatic computer aided quantification of synovitis using dynamic MRI and the impact of movement correction on signal to noise (SNR) and region of interest (ROI) analysis. Abstract for American Congress of Rheumatology 2009. 31 Kubassova O, Boesen M, Cimmino MA et al. A computeraided detection system for rheumatoid arthritis MRI data interpretation and quantification of synovial activity. Eur J Radiol 2009. 32 Axelsen MB, Poggenborg RP, Kubassova O et al. Reliability and responsiveness of dynamic contrast

www.rheumatology.oxfordjournals.org

Computer-aided analysis of dynamic gadolinium-enhanced MRI

enhanced magnetic resonance imaging in arthritis. Ann Rheum Dis 69(Suppl. 3):348. 33 Arnett FC, Edworthy SM, Bloch DA et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988; 31:315–24.

www.rheumatology.oxfordjournals.org

34 Haavardsholm EA, Ostergaard M, Hammer HB et al. Monitoring anti-TNFalpha treatment in rheumatoid arthritis: responsiveness of magnetic resonance imaging and ultrasonography of the dominant wrist joint compared with conventional measures of disease activity and structural damage. Ann Rheum Dis 2009;68:1572–9.

143