Use magnetic resonance imaging to assess articular cartilage

Abstract

Magnetic resonance imaging (MRI) enables a noninvasive, three-dimensional assessment of the entire joint, simultaneously allowing the direct visualization of articular cartilage. Thus, MRI has become the imaging modality of choice in both clinical and research settings of musculoskeletal diseases, particular for osteoarthritis (OA). Although radiography, the current gold standard for the assessment of OA, has had recent significant technical advances, radiographic methods have significant limitations when used to measure disease progression. MRI allows accurate and reliable assessment of articular cartilage which is sensitive to change, providing the opportunity to better examine and understand preclinical and very subtle early abnormalities in articular cartilage, prior to the onset of radiographic disease. MRI enables quantitative (cartilage volume and thickness) and semiquantitative assessment of articular cartilage morphology, and quantitative assessment of cartilage matrix composition. Cartilage volume and defects have demonstrated adequate validity, accuracy, reliability and sensitivity to change. They are correlated to radiographic changes and clinical outcomes such as pain and joint replacement. Measures of cartilage matrix composition show promise as they seem to relate to cartilage morphology and symptoms. MRI-derived cartilage measurements provide a useful tool for exploring the effect of modifiable factors on articular cartilage prior to clinical disease and identifying the potential preventive strategies. MRI represents a useful approach to monitoring the natural history of OA and evaluating the effect of therapeutic agents. MRI assessment of articular cartilage has tremendous potential for large-scale epidemiological studies of OA progression, and for clinical trials of treatment response to disease-modifying OA drugs.

Introduction

Magnetic resonance imaging (MRI) provides high-spatial-resolution, multiplanar imaging and excellent tissue contrast. This enables a three-dimensional assessment of all components of the joint simultaneously, allowing direct visualization of articular cartilage. With the advances in techniques and development of dedicated sequences, MRI has become the imaging modality of choice in both clinical and research settings of musculoskeletal diseases, in particular osteoarthritis (OA), a chronic joint disease characterized by destruction and progressive loss of articular cartilage and clinical symptoms including pain, stiffness and impaired function.

Although OA is a major clinical and public health problem resulting in substantial burden to the individuals and society, the pathogenesis of OA has not been fully understood. Consequently, limited preventive strategies have been identified to modify the predisposing factors of OA, and no current treatments have been shown to slow disease progression, with end-stage symptomatic disease treated by joint replacement surgery. An accurate assessment of structural changes in articular cartilage is required to monitor the progression of OA and evaluating the therapeutic response. The understanding of OA and the development of therapeutic interventions have been hampered by the lack of validated, noninvasive tools to quantify articular cartilage and assess the severity and progression of the disease. Early identification of disease has the potential to provide novel opportunities to trial both nonpharmacological and pharmacological (e.g. disease-modifying osteoarthritis drugs [DMOADs]) agents to help curtail the burden of this prevalent and debilitating disease.

The current gold standard for the assessment of OA in clinical and epidemiological settings is based on radiographs. This assumes that joint space narrowing or width may be used as a surrogate for articular cartilage. Although substantial advances have been made with radiographic techniques, the radiographic methods have significant limitations when used to measure disease severity and progression. Radiographs provide two-dimensional images of a three-dimensional structure, and only provide direct assessment of bony features. Radiographs are unable to delineate articular cartilage directly and thus are unable to detect early cartilage pathologies. Once radiographic changes are detected, significant disease is already present: with grade 1 joint space narrowing being detected, 11–13% of cartilage has been lost [Jones et al. 2004]. The measurement of joint space narrowing is prone to measurement error related to positioning of the joint, and its position respective to radiograph and source of radiation [Altman et al. 1996; Buckland-Wright et al. 1995; Ravaud et al. 1996]. This is of particular concern in multicentre studies and longitudinal studies. Radiographic grading provides a crude and insensitive method by which to assess disease progression, thus a larger sample size followed up for a prolonged period is required for studies examining radiographic progression of OA. Moreover, structures other than articular cartilage in the joint space may result in changes in joint space narrowing: meniscal extrusion has been shown to contribute to joint space narrowing independent of cartilage status [Adams et al. 1999; Gale et al. 1999].

There has been increasing interest in using MRI to examine joint cartilage as a measure of disease severity and progression of OA. In 1996, an expert consensus concluded that for knee OA, MRI would likely be useful, once validated methods of measurement are developed to measure articular cartilage [Altman et al. 1996]. Over the past decade, much work has been done in this area. MRI enables quantitative (cartilage volume and thickness) and semiquantitative assessment of articular cartilage, and recently the assessment of cartilage matrix composition, such as T1 relaxation time in rotating frame (T1rho), T2 relaxation time, and delayed gadolinium-enhanced MRI of cartilage (dGEMRIC), thus providing new insight into the natural history and determinants of cartilage pathologies. The emerging evidence suggests that MRI is able to show early morphological changes in articular cartilage related to OA, prior to the onset of radiographic disease. Evidence from MRI studies has suggested that there is a continuum from a healthy joint, through to the preclinical state, and ultimately to the changes of end-stage OA, at which time joint replacement is indicated. MRI provides accurate and reliable assessment of articular cartilage which is sensitive to change [Eckstein et al. 2006]. This suggests that using MRI cartilage morphology measures as the end point will reduce the number of participants needed in clinical trials and reduce the length of clinical trials to detect the treatment response to DMOADs with sufficient statistical power, having the potential to decrease the cost of clinical trials.

Quantitative assessment of articular cartilage

Cartilage volume

Cartilage is a three-dimensional structure. Cartilage volume can be measured from MR images [Cicuttini et al. 2000; Peterfy et al. 1994]. This has the theoretical benefit of being less subject to repositioning error than joint space width obtained from plain radiography, since cartilage is measured in three dimensions, as well as removing the effect of meniscal extrusion on apparent changes in joint space narrowing [Adams et al. 1999; Gale et al. 1999]. Although important in cross-sectional studies, this becomes more critical in longitudinal studies, where repositioning error will have a greater impact on study results.

Validation of cartilage volume measurement

Before using cartilage volume as an endpoint in studies, it requires validation: verification that cartilage measurement using MRI can measure cartilage volume accurately and reliably. This involves comparing the new measurement with a gold standard.

The face validity (i.e. the extent to which the MRI-derived results measure articular cartilage) of the measurement of cartilage volume have been established by a number of studies, and the degree of accuracy has been shown to be similar for all cartilage surfaces of the knee [Burgkart et al. 2001; Cicuttini et al. 1999, 2000; Dupuy et al. 1996; Eckstein et al. 1998a, 1998b; Graichen et al. 2004; Peterfy et al. 1994; Pilch et al. 1994; Piplani et al. 1996; Sittek et al. 1996]. These studies established that MRI-derived cartilage volume measurement had high face validity using cadaveric and surgical specimens [Burgkart et al. 2001; Cicuttini et al. 1999, 2000; Dupuy et al. 1996; Eckstein et al. 1998b; Graichen et al. 2004; Peterfy et al. 1994; Pilch et al. 1994; Piplani et al. 1996; Sittek et al. 1996] and alternative imaging methods (CT arthrography) [Eckstein et al. 1998a], with random errors of 5–10%. There is no consistent or systematic underestimation or overestimation of cartilage volume. Thus, MRI-derived measurements of cartilage volume appear to be a valid method to measure knee cartilage. The resolution, signal-to-noise ratio and contrast-to-noise ratio are substantially improved using a 3.0-T imaging system, resulting in higher accuracy in quantifying cartilage volume compared with a 1.5-T system [Bauer et al. 2006; Eckstein et al. 2005].

Reproducibility of cartilage volume measurement

To be a useful tool, cartilage volume not only has to be valid, but also be reproducible. Reproducibility is the extent to which measurement of the phenomenon of interest provides the same results upon repeat applications under the same circumstances. This may be tested using the test–retest method. It can be assessed for both intra-observer (intrarater) reliability, where one observer measures disease on the same image on different occasions, and inter-observer (interrater) reliability, where more than one observer measure disease at different points in time. This measures the extent to which the test is free of random error (subject variation, instrument variation and observer variation). MRI-derived measurement of knee cartilage volume has been assessed by many investigators in different compartments of the knee and shown to have high reproducibility [Cicuttini et al. 1999, 2000; Dupuy et al. 1996; Eckstein et al. 1998b; Marshall et al. 1995b; Peterfy et al. 1994; Piplani et al. 1996; Sittek et al. 1996]. The interobserver reproducibility ranges from 0.4% to 7.8% (coefficient of variation [CV]); the intra-observer reproducibility from 0.3% to 6.4% (CV), and the image re-image reproducibility from 1.1% to 9.7% (CV). Raynauld and colleagues reported high reproducibility of MRI cartilage volume measurement, by assessing the inter-reader agreement, the test–retest reliability, and the patient positioning reliability on 48 MR examinations of the knee from normal subjects and patients with different stages of symptomatic knee OA [Raynauld et al. 2003]. The interreader agreement of measurements was excellent, as shown by intraclass correlation coefficients (ICCs) ranging from 0.958 to 0.997 for global cartilage, 0.974 to 0.998 for the compartments and 0.943 to 0.999 for the femoral condyles. The test–retest reliability of intrareader agreement was excellent, with Pearson correlation coefficients ranging from 0.978 to 0.999. The patient positioning reliability was also excellent, with Pearson correlation coefficients ranging from 0.978 to 0.999 [Raynauld et al. 2003]. Moreover, cartilage volume measured at subregions of the joint surface may reduce the precision errors relative to those from measuring the entire cartilage plate [Kshirsagar et al. 1998]. Cartilage volume measurement using a 3.0-T imaging system tends be more reproducible than at 1.5-T system and may therefore provide superior ability to detect changes in cartilage status over time and to determine responses to treatment with structure-modifying drugs [Eckstein et al. 2005].

Sensitivity to change of cartilage volume measurement

Although articular cartilage can be seen with proper acquisition sequences, the quantification of cartilage volume change has been a challenge due mainly to postscan processing times. In OA, loss of cartilage volume may be accelerated, being a multifactorial phenomenon which may precede, accompany and/or result from other structural changes within the joint.

Eckstein and colleagues analysed the long-term and resegmentation precision of quantitative cartilage measurement. They found scanner drift and changes in imaging or patient conditions were not a critical problem in quantitative cartilage analysis [Eckstein et al. 2002]. The study suggested that in longitudinal studies, image analysis of sequential data should be performed within the same postprocessing session. They concluded that under these conditions, quantitative cartilage measurement is a very powerful method to assess structural cartilage changes in OA [Eckstein et al. 2002]. MRI also provides a far more sensitive measure of change in cartilage volume than the joint radiograph, with 11–13% of cartilage volume being lost before the first changes of radiographic joint space narrowing can be detected [Jones et al. 2004]. Several studies have reported data on the change in cartilage volume in longitudinal studies, showing that measurement of knee cartilage volume is sensitive to change in both normal subjects [Hanna et al. 2005; Wluka et al. 2004a] and those with OA or at high risk of OA [Cicuttini et al. 2002a, 2002b; Raynauld et al. 2004; Wluka et al. 2002b]. In healthy postmenopausal women without clinical knee OA, tibial cartilage volume decreases on average by 2.4% per year [Wluka et al. 2004a]. In healthy middle-aged men, tibial cartilage volume reduces by 2.8% per year [Hanna et al. 2005]. Wluka and colleagues reported an annual tibial cartilage volume loss rate of 5.3% in patients with symptomatic and radiographic knee OA [Wluka et al. 2002b]. With the same cohort, Cicuttini and colleagues reported the higher patellar cartilage volume loss rate in women (5.3%) compared with men (3.5%) [Cicuttini et al. 2002a]. Raynauld and colleagues reported cartilage volume loss rate (period estimate) at a series of follow-up time points: 3.8% for global cartilage volume loss and 4.3% for medial compartment cartilage volume loss at 6 months, 3.6% and 4.2% at 12 months, and 6.1% and 7.6% at 24 months. Furthermore, discriminant function analysis identified 2 groups of patients, 21 patients progressed slowly (<2% of global cartilage volume loss) and 11 patients progressed rapidly (>15% of global cartilage volume loss), over the 2 years of study [Raynauld et al. 2004]. This study detected cartilage volume loss in the absence of changes in radiographic joint space narrowing, suggesting MRI is more sensitive to change than radiograph [Raynauld et al. 2004]. Cicuttini and colleagues reported a rate of tibial cartilage volume loss of 4.1% per year in a sample of participants undergoing partial meniscectomy [Cicuttini et al. 2002b].

Cartilage thickness

Whilst many authors have examined volume as an outcome measure, others have investigated cartilage thickness [Cohen et al. 1999; Eckstein et al. 1996, 1997, 1998a; Graichen et al. 2004; Kladny et al. 1996; Marshall et al. 1995a; Sittek et al. 1996]. Although this may appear an attractive measure, being a one-dimensional estimate of a three-dimensional structure, it shares one of the shortcomings of the use of joint space width as a measure of disease severity in that it may be more prone to repositioning error than cartilage volume. It is also sensitive to errors related to partial volume averaging, when a single measurement is used. Theoretically, this measure has the problem of ensuring the same thickness is measured on subsequent occasions. The method of coregistration of images may be used to ensure that this occurs. This adds significantly to the technical expertise and cost required for this measurement.

Validation of cartilage thickness measurement

Cartilage thickness has been investigated as an alternative and time-saving measure of articular cartilage. Although there are methodological issues including the difficulty of defining the section to be measured, it has been established that MRI-derived measurements of cartilage thickness and cartilage thickness pattern had high validity using cadaveric and surgical specimens and alternative imaging methods (CT arthrogram or stereophotogrammetry), comparable with cartilage volume [Cohen et al. 1999; Eckstein et al. 1996, 1997, 1998a; Graichen et al. 2004; Kladny et al. 1996; Marshall et al. 1995a; Sittek et al. 1996]. In quantifying cartilage thickness, central and weight-bearing regions of the femoral condyles can provide more accurate measurement than boundary and non-weight-bearing regions [Koo et al. 2005]. MRI-derived measurements of cartilage thickness appear to be a valid method to measure knee cartilage. Cartilage thickness measurement using a 3.0-T imaging system showed nonsignificant difference from measurements using a 1.5-T system [Eckstein et al. 2005; Kornaat et al. 2005b].

Reproducibility of cartilage thickness measurement

One centre has attempted reproducibility studies of cartilage thickness [Eckstein et al. 1996, 1998a, 1998b]. This has used highly sophisticated mapping to measure thickness profiles. It has not been tested longitudinally, or with coregistration, to ensure cartilage thickness at the same position were compared. This is a potential pitfall when cartilage thickness is used as an outcome measure in longitudinal studies. The results of these studies suggest misclassification of thickness occurs in approximately 33% of cases, in all studies and cartilages. A second centre performed intra-observer and interobserver reproducibility studies [Karvonen et al. 1990]. This group reported an intra-observer variability of 3.2–10.5% (CV). Interobserver variability ranged between 5.4–29.7% (CV). The reproducibility of this measure is less than cartilage volume. Whilst some early investigators used this measure alone [Karvonen et al. 1990, 1994], more recent studies use thickness as an outcome measure in addition to volume measurements.

Sensitivity to change of cartilage thickness measurement

Cartilage thickness naturally decreases with normal aging, with an annual reduction of 0.3–0.5% [Hudelmaier et al. 2001, 2003]. In young adults with acute anterior cruciate ligament injury, over 2 years, significant cartilage thinning is observed in the trochlea of the femur at a rate of 4.3%, while significant cartilage thickening is observed in the central medial aspect of the femur at a rate of 2.7% [Frobell, 2011]. In patients with radiographic knee OA, over 12 months, the mean cartilage thickness reduction in the central subregion of medial femoral condyle is 12 µm in knees without pain, 27 µm in those with infrequent pain and 54 µm in those with frequent pain [Eckstein et al. 2011]. In a subsample of the OA Initiative progression subcohort, the mean cartilage thickness reduces by 2.8% in the central aspect of medial femoral condyle, 1.9% in the total medial femoral condyle, 0.9% in the central medial tibia and 0.5% for the entire medial tibia [Wirth et al. 2009]. In patients with radiographic knee OA, the total tibiofemoral cartilage thickness reduces by 0.8–1.3% in the first year and by 0.7–0.8% in the second year [Wirth et al. 2011]. There is data suggesting that the mean cartilage thickness over the entire subchondral bone area tend to be more reproducible and more sensitive to change than cartilage thickness over the cartilaginous area or maximum cartilage thickness [Hudelmaier et al. 2010]. Cartilage thickness has shown similar sensitivity to assess cartilage loss compared with cartilage volume in a clinical trial [Raynauld et al. 2008].

Subregional measurement of cartilage volume and thickness

There has been interest in measuring cartilage volume and thickness in subregions of the femoral and tibial cartilage plates. In a longitudinal study of 107 patients with knee OA performed by Pelletier and colleagues subregion analysis revealed that the greatest cartilage volume loss at 24 months was found in the central area of the medial tibial plateau (15%) and of the medial femoral condyle (12%) [Pelletier et al. 2007]. Recent studies demonstrated that the greatest rate of cartilage volume and thickness loss was observed in the central medial femoral condyle [Eckstein et al. 2009; Hunter et al. 2009; Wirth et al. 2009]. However, in line with other investigation [Raynauld et al. 2008], the sensitivity to change in the subregions is not consistently higher than in the total cartilage plates across studies, owing to the high standard deviation of the subregional measurements. Subregional assessment of cartilage may have missed the effect of intervention [Raynauld et al. 2008]. Thus, the potential of gaining statistical power with the use of cartilage volume and thickness change in knee subregions as an outcome would be negated by high interpatient variability [Raynauld et al. 2008]. The spatial heterogeneity of cartilage loss in OA patients may provide an explanation why subregional assessment of cartilage does not show a higher sensitivity to change. There is evidence that knee alignment determines the patterns of cartilage loss in the medial versus lateral compartment [Eckstein et al. 2008; Sharma et al. 2008] and that local biomechanical environment may be responsible for certain subregions to progress faster than others in certain subjects [Roemer et al. 2009]. Furthermore, in contrast to weight-bearing radiographs, MRI is non-weight-bearing, thus the central regions on knee MRIs may not be anatomical and may not be the weight-bearing regions particularly when there is malalignment.

Shortcomings of current methods of quantifying articular cartilage

Neither cartilage volume nor thickness measurements can assess the quality of cartilage present. This has been raised as a significant concern in cross-sectional studies, since theoretically, and in animal models, cartilage change in OA has been shown to be biphasic, with cartilage swelling occurring early, prior to cartilage thinning and loss [Calvo et al. 2001; Watson et al. 1996]. Histopathologically, the swelling of cartilage detected by MRI correlates with depletion in matrix proteoglycans and cellular loss, which are closely related to the progression of OA at the earliest stages [Calvo et al. 2004]. Longitudinal studies are less prone to this bias as swollen cartilage will have faster rates of loss. However, attempts to date have been unable to correlate cartilage quality with its appearance on MRI.

Semiquantitative assessment of articular cartilage

A number of scoring methods have been used to quantify regional cartilage loss, identifying irregularities of the cartilage. One such scoring method is that of cartilage defects. Cartilage defects are identifiable using MRI where the cartilage surface or the cartilage adjacent to bone is irregular, or where there is loss of cartilage thickness [Ding et al. 2005c]. Histologically, these lesions are highly correlated with the Mankin scale for grading cartilage [McGibbon and Trahan, 2003].

Validity of cartilage defects measurement

Several studies have examined the diagnostic performance (sensitivity and specificity) of MRI-detected cartilage lesions using different acquisition sequences against surgically confirmed cartilage lesions and arthroscopy [Bredella et al. 1999; Disler et al. 1996; Duc et al. 2007, 2008; Kawahara et al. 1998; Kijowski et al. 2009, 2010; Murphy, 2001; Wong et al. 2009; Yoshioka et al. 2004]. The diagnostic performance is highest on patellar cartilage and lowest on lateral tibial cartilage, and is higher for deep cartilage lesions than for superficial lesions. Using a 1.5-T imaging system, the sensitivity ranges between 40% and 94%, and the specificity ranges between 70% and 100%. With a 3.0-T imaging system, the sensitivity ranges between 66% and 80%, and the specificity ranging between 80% and 97%. Although some studies have shown improved diagnostic performance of 3.0-T MRI protocol for evaluating cartilage defects compared with 1.5-T protocol [Fischbach et al. 2005; Kijowski et al. 2009; Link et al. 2006; Masi et al. 2005; Wong et al. 2009], there are significant overlaps in the sensitivity and specificity between 3.0-T and 1.5-T MRI protocols and some differences are not significant [Kijowski et al. 2009; Link et al. 2006].

Various MRI pulse sequences have been used for the assessment of cartilage defects, which has led to debates as per the choice of appropriate pulse sequences for MRI-based semiquantitative assessment of cartilage defects in OA research [Hayashi et al. 2010]. Similar overall diagnostic performance (relatively high sensitivity and specificity) for the detection of focal cartilage defects have been shown for fat-suppressed T1-weighted spoiled gradient recalled acquisition at steady state (SPGR) [Disler et al. 1996; Recht et al. 1996; Yoshioka et al. 2004], fat-suppressed T2-weighted fast spin echo (FSE) [Bredella et al. 1999; Broderick et al. 1994; Kawahara et al. 1998; Yoshioka et al. 2004], selective water-excitation SPGR [Mohr et al. 2003], dual echo steady state (DESS) [Woertler et al. 2000], and fat-suppressed driven equilibrium Fourier transform (DEFT) [Yoshioka et al. 2004]. Although SPGR sequences provide high spatial resolution, these sequences are vulnerable to artefacts (motion artefacts due to long imaging time and susceptibility artefacts). These should be considered after previous surgery, in particular after cartilage repair procedures [Link et al. 2007]. Despite the assessment of cartilage defects using SPGR sequences which may entail additional measurement error and misclassification due to these vulnerabilities, cartilage defects measured using these sequences have been associated with clinically important outcomes such as pain, radiographic OA, cartilage volume loss, cartilage breakdown markers, and knee replacement surgery in multiple cross-sectional and longitudinal studies [Ding et al. 2007b].

Reproducibility of cartilage defects measurement

For the assessment of cartilage defects, the intra-observer reliability (expressed as ICC) was 0.89–0.90, and the interobserver reliability (expressed as ICC) was 0.85–0.90 [Ding et al. 2005c]. For the assessment of cartilage lesions using Whole-Organ MRI Score, the interobserver agreement (expressed as ICC) was >0.98 [Peterfy et al. 2004]. The intra-observer reproducibility (expressed as ICC) of another scoring system, the Knee OA Scoring System, was 0.76–0.96, and the interobserver reproducibility was 0.63–0.91 [Kornaat et al. 2005a].

Sensitivity to change of cartilage defects measurement

In a study of 43 patients over 1–5 years (mean 1.8 years), among the cartilage lesions present at baseline, 12% were not detectable at follow up, 6% regressed to a lower grade, 32% remained the same grade and 50% progressed to a higher grade. There were 84 new lesions during the study period [Biswal et al. 2002]. In a study of 86 participants without knee OA, Wang and colleagues reported that over 2 years the cartilage defect score increased, and that the medial tibiofemoral, lateral tibiofemoral and patellar cartilage defect score increased in 63%, 65% and 36% of the subjects, remained unchanged in 32%, 30% and 46% of the subjects and decreased in 5%, 5% and 18% of the subjects, respectively [Wang et al. 2006]. In a convenience midlife sample of 325 subjects, over 2.3 years, patellar cartilage defect score increased significantly, 12%, 13% and 22% of subjects had an increase in cartilage defect scores, whereas 13%, 12% and 13% of subjects had a decrease in cartilage defect scores at the medial, lateral and patellar sites, respectively [Ding et al. 2005a]. Overall, 33% of the subjects had a worsening (≥1 point increase) and 37% of the subjects had an improvement (≥1 point decrease) in cartilage defect score in any knee compartment during 2.3 years [Ding et al. 2006]. A study of symptomatic OA patients found increased cartilage defect score over the 2 years, and the heterogeneity with 81% progressing, 15% remaining stable and 4% regressing [Davies-Tuck et al. 2008]. It may be that the improvement in cartilage defects is, at least in part, attributable to measurement issues, however since cartilage defects had to be present on ≥2 consecutive slices [Ding et al. 2005c], it seems unlikely that this is due to measurement error alone in multiple independently assessed cohorts. These data suggest that there may be some repair of articular cartilage [Buckwalter and Mankin, 1998; Caplan et al. 1997] which decreases with increasing age.

Assessment of cartilage matrix composition

Although MRI-derived quantitative and semiquantitative assessments of cartilage morphology have provided a useful tool to monitor cartilage loss and disease progression in OA, there is significant interest in using MRI to detect very early cartilage changes, prior to the morphological changes. Cartilage matrix composition can be assessed by MRI using several techniques, including T1rho, T2 relaxation time and dGEMRIC, which have been shown to be sensitive to early cartilage damage and a potential early marker for degenerative joint disease.

Using animal models and cartilage explants, T1rho is strongly correlated with changes in proteoglycan content [Akella et al. 2001; Regatte et al. 2003; Wheaton et al. 2004, 2005], and increase in T1rho is correlated with proteoglycan loss [Duvvuri et al. 2002], suggesting T1rho could be used as a method of quantifying proteoglycan changes in early stage OA. In vivo reproducibility of average T1rho of patellar cartilage was found to be 5% (CV) [Pakin et al. 2006]. In anterior cruciate ligament-injured knees, T1rho significantly increased in the posterolateral tibia and persisted 1 year after anterior cruciate ligament reconstruction, and T1rho of weight-bearing medial tibiofemoral cartilage was significantly elevated at 1 year after the surgery [Li et al. 2011]. Patients with OA patients showed significant elevation in T1rho compared with asymptomatic controls, and the T1rho was associated with the degree of cartilage degeneration [Li et al. 2005; Regatte et al. 2004]. These findings suggest that T1rho may be a promising clinical tool for OA detection and treatment monitoring.

In vivo, increasing T2 relaxation time is associated with the distribution of cartilage water and it is sensitive to small water content changes [Liess et al. 2002]. Aging has been shown to be associated with an asymptomatic increase in T2 of the transitional zone of articular cartilage. The diffuse increase in T2 in senescent cartilage is different in appearance than the focally increased T2 observed in damaged articular cartilage [Mosher et al. 2000]. When subjects with OA were compared with those without OA, femoral and tibial cartilage T2 values were higher in those with OA compared with healthy subjects [Dunn et al. 2004; Stahl et al. 2007]. However, no significant difference in T2 value was seen when those with mild and severe OA were compared [Dunn et al. 2004]. Similar results were shown in another study that found no differences in T2 values across the Kellgren–Lawrence grades of OA [Koff et al. 2007]. A longitudinal study examining change in women with and without OA showed no significant change in T2 values over 1 year in either group [Stahl et al. 2007]. These results indicate whilst T2 relaxation time differs in those with and without OA, it is not a sensitive marker of radiographic severity thus does not help to differentiate between mild and more severe OA. However, in contrast, in a group with and without OA, T2 values correlated with both clinical symptoms and cartilage morphology (cartilage volume and thickness), particularly in the medial compartment [Dunn et al. 2004]. This suggestion of clinical significance warrants further investigation.

Some studies compared the sensitivity of T1rho and T2 relaxation time. In a study of 10 healthy volunteers and 9 OA patients, a significant difference was observed in T1rho within patellar and femoral cartilage between the 2 groups. Although a significant correlation was found between T1rho and T2 relaxation time, the difference of T2 was not significant between controls and OA patients [Li et al. 2005]. Another study also found T1rho was superior to T2 in differentiating early OA from normal cartilage, although both T1rho and T2 values were significantly higher in early OA patients compared with healthy subjects [Stahl et al. 2009]. In an animal model, while T2 relaxation rates varied randomly without any particular trend, yielding a poor correlation with sequential depletion of proteoglycan, there was excellent linear correlation between the percentage of proteoglycan and the T1rho relaxation rate, suggesting T1rho-weighted imaging, rather than T2-weighted imaging, is sensitive to sequential depletion of proteoglycan in bovine cartilage and can be used to quantify proteoglycan-induced changes [Regatte et al. 2002]. In cartilage specimens obtained from total knee replacement surgery with various clinical grades of OA, all cartilage specimens showed elevated T1rho and T2 relaxation times compared with healthy cartilage tissue. T1rho showed a higher dynamic range (>100%), and was correlated with the degree of cartilage degeneration. However, T2 relaxation times showed lower dynamic range [Regatte et al. 2006]. These data suggest that T1rho relaxation mapping is a sensitive noninvasive marker for quantitatively predicting and monitoring the status of macromolecules in early OA. The higher dynamic range of T1rho can be exploited to measure even small macromolecular changes with greater accuracy compared with T2 [Regatte et al. 2006].

dGEMRIC involves the intravenous injection of MRI contrast agent and the acquisition of a T1 map to allow a quantitative assessment of the glycosaminoglycan content within the cartilage, where the contrast agent is distributed in higher concentration in areas with lower glycosaminoglycan content. After application of gadolinium, the variations in T1 reflected about 50% variations in glycosaminoglycan, suggesting that dGEMRIC has potential for monitoring glycosaminoglycan content of cartilage in vivo [Bashir et al. 1997]. The dGEMRIC measurement of glycosaminoglycan has been validated against the reference standards of histological and biochemical measurement of glycosaminoglycan content [Bashir et al. 1999; Trattnig et al. 1999]. The T1 measurement in dGEMRIC is referred to as the dGEMRIC index. In 31 patients with knee OA, compartments without joint space narrowing had a higher dGEMRIC index than those with any level of narrowing. In knees with one unnarrowed (spared) and one narrowed (diseased) compartment, the dGEMRIC index was greater in the spared versus the diseased compartment. In spared compartments, there was a trend toward a lower dGEMRIC index with increasing Kellgren–Lawrence radiographic severity grade. There was a range of dGEMRIC values in the spared compartments within a given Kellgren–Lawrence grade, demonstrating biochemical differentiation of disease in radiographically comparable compartments. Valgus-aligned knees tended to have lower dGEMRIC values laterally and varus-aligned knees tended to have lower dGEMRIC values medially; as a continuous variable, alignment correlated with the lateral:medial dGEMRIC ratio. These data suggest the biochemical information provided by dGEMRIC scans may improve the differentiation of disease status within a given radiographic grade, especially in early OA [Williams et al. 2005]. Importantly, recent data showed low dGEMRIC index at baseline was predictive of the development of knee OA, suggesting the role of dGEMRIC index as a clinically relevant measure of cartilage integrity [Owman et al. 2008].

However, data is limited on the longitudinal assessment of these compositional measures of articular cartilage. Although these compositional measures may play an important role in detecting early cartilage changes before cartilage defects occur, the natural course of these measures and their relationship with cartilage morphological changes and clinical outcomes warrant further investigation.

Association between MRI cartilage measures and radiographic features

Since radiographic joint space narrowing is used as a surrogate measure of articular cartilage, it would be expected that MRI-derived cartilage assessments are associated with radiographic features. However, the evidence has been conflicting. Plain radiography is insensitive to early changes such as the development of cartilage defects and early cartilage loss that are unequivocally occurring prior to any radiographic diminution of the joint space. Whilst joint space narrowing and joint space width have been related to cartilage volume, longitudinal studies suggest that changes in joint space narrowing and joint space width are less sensitive in detecting cartilage loss than MRI [Raynauld et al. 2004, 2006].

A cross-sectional study by Cicuttini and colleagues showed a strong negative, linear association between medial and lateral tibial cartilage volume and increasing grade of joint space narrowing, and that the inverse relation was even stronger when adjusting for age, gender and body mass index [Cicuttini et al. 2003b]. In a 1-year longitudinal study, Bruyere and colleagues reported a moderate but significant association between changes in joint space narrowing and changes in cartilage volume and thickness patients with knee OA [Bruyere et al. 2007]. Recently Pelletier and colleagues found, in patients with knee OA over 2 years, there was a significant correlation between reduction in joint space width and cartilage volume loss in the central weight-bearing area of the condyles and the plateaus as well as in the medial compartment [Pelletier et al. 2007]. Cartilage defects have been shown to be significantly associated with radiographic features of OA [Boegard et al. 1998; Brandt et al. 1991; Link et al. 2003]. In a study of 224 patients with symptomatic knee OA, cartilage loss assessed by semiquantitative MRI grading score was associated with joint space narrowing [Amin et al. 2005]. Kellgren–Lawrence score has been shown to be correlated with the grade of cartilage lesions [Link et al. 2003].

However, the study by Gandy and colleagues observed a 0.21 mm reduction in joint space width in weight-bearing extended radiographs found no significant cartilage volume change in any of the knee compartments [Gandy et al. 2002]. Raynauld and colleagues found no significant change in weight-bearing semiflexed positioned radiographs in OA patients over 2 years, but reported significant cartilage volume change in the medial and lateral femorotibial compartment. There was no significant correlation between baseline joint width and cartilage volume in the medial compartment, and between cartilage volume loss and radiographic changes over 2 years [Raynauld et al. 2004].

Fractal analysis of trabecular bone texture on radiographs has been shown to be a promising approach for assessing early knee OA [Kraus et al. 2009; Wolski et al. 2010, 2011]. In subjects without radiographic knee OA, trabecular bone texture in the proximal tibia is different between those with and without tibiofemoral cartilage defects, where trabecular bone texture roughness is greater in those with cartilage defects than in those without [Wolski et al. 2011]. These findings suggest a potential role of MRI cartilage assessment in studies evaluating new radiographic features.

Association between MRI cartilage measures and clinically important outcomes

Cartilage measures and symptoms

A number of MRI studies have demonstrated significant relationships between cartilage structural changes and knee pain. Cartilage volume loss has been associated with WOMAC pain changes over 24 months in patients with knee OA [Raynauld et al. 2006]. In a cross-sectional study of 133 postmenopausal females, alterations in patella volume were associated with pain, function and global scores of the WOMAC. In participants with more knee pain, there was an association with reduced patella cartilage volume [Hunter et al. 2003]. A longitudinal study of 117 subjects with knee OA over 2 years demonstrated a relationship between baseline cartilage volume and knee symptoms (WOMAC pain, stiffness and function scores), and a correlation between the loss of cartilage volume over 24 months and the worsening of symptoms [Wluka et al. 2004b]. This study suggested that although cartilage is not a major determinant of symptoms in knee OA, it does relate to symptoms [Wluka et al. 2004b]. One study found in patients with OA the grades of cartilage lesions rated from MRI were associated with clinical symptoms including pain, stiffness and limited function assessed by WOMAC scores [Link et al. 2003]. Full-thickness cartilage defects accompanied by subchondral bone defects are associated with pain in OA [Sowers et al. 2003]. The correlation between knee pain and cartilage defects has been reported in particular when the defect is moderate to severe (grade 2–3) [Zhai et al. 2006, 2007].

However, there are a few studies which found no associations between longitudinal cartilage change and worsening of symptoms as determined by WOMAC score [Phan et al. 2006; Raynauld et al. 2004]. Zhai and colleagues found in both younger and older adults that cartilage volume was not associated with knee pain [Zhai et al. 2006, 2007]. These data suggest that pain in knee OA may arise from a multitude of factors and is likely to originate from structures other than cartilage that are innervated by nociceptors including subchondral bone, meniscus, synovium, capsule or ligaments.

Cartilage measures and joint replacement

Joint replacement is considered a surrogate marker of end-stage symptomatic OA. In those with symptomatic knee OA, a higher cartilage defect score is associated with a sixfold increased risk of knee joint replacement over 4 years [Wluka et al. 2005]. Similarly, the rate of tibial cartilage volume loss over 2 years is an independent predictor of joint replacement, reporting a 20% increase in risk of undergoing joint replacement over 4 years for every 1% increase in the rate of cartilage volume loss. When those in the top tertile of tibial cartilage volume loss were compared with those in the bottom tertile, they had a 7.1-fold increased risk of knee replacement [Cicuttini et al. 2004].

Role of MRI cartilage measures in understanding risk factors for OA and prevention of OA

The increased knowledge of the significance of different structural changes in OA has enabled better understanding of the risk factors for OA and novel approaches to preventing OA. This has been made possible by the use of sensitive measures of articular cartilage, enabling the examination of early cartilage alterations prior to the presence of radiographic disease.

Obesity

Obesity has been recognized as the most important modifiable risk factor for OA [Felson et al. 2000]. Obesity has been shown to influence MRI-assessable features of OA, which may in turn affect the risk of knee OA. Emerging evidence suggests that the means by which obesity affects joint health is likely to involve both biomechanical and metabolic mechanisms. Biomechanically, obesity increases the risk of OA directly through the increasing load on joints [Felson et al. 2000]. Metabolic factors are the product of adipose tissue, now considered an endocrine organ [Calabro and Yeh, 2007]. Adipose tissues release many factors, including pro-inflammatory cytokines such as tumour necrosis factor (TNF-α) and interleukin-6, and adipokines, such as leptin, adiponectin and resistin [Pottie et al. 2006]. Both TNF-α and interleukin-6 have been implicated in cartilage destruction in OA [Evans, 2005; Malemud, 2004], while leptin is a key regulator of chondrocyte metabolism and plays an important role in the pathophysiology of OA [Dumond et al. 2003]. Chondrocytes from joint cartilage have been shown to express leptin receptors [Dumond et al. 2003; Figenschau et al. 2001], which, when stimulated, produce nitric oxide, leading to inflammatory alterations in cartilage including phenotype loss of chondrocytes, apoptosis and metalloproteases activation [Kim et al. 2002; Otero et al. 2003].

Obesity, assessed by increase weight and body mass index, has been shown to be associated with reduced cartilage volume [Cicuttini et al. 2003a; Ding et al. 2003; Nishimura et al. 2005; Raynauld et al. 2004; Teichtahl et al. 2009b] but more consistently with increased cartilage defects [Anandacoomarasamy et al. 2009; Ding et al. 2005b, 2008; Hanna et al. 2007; Teichtahl et al. 2008]. However, body mass index as a measure of obesity, does not account for the pattern of fat distribution or body composition: it does not discriminate adipose from nonadipose body mass [Roubenoff, 1996]. Studies on body composition and knee cartilage provide new insight into the mechanism by which obesity is associated with an increased risk of knee OA. In a longitudinal study of healthy, middle-aged adults without knee OA, loss of muscle mass was associated with an increased loss of medial and lateral tibial cartilage volume over 2 years [Cicuttini et al. 2005]. A cross-sectional study of 297 adults with no clinical knee OA found that greater fat-free mass was associated with increased cartilage volume, while greater fat mass was associated with reduced cartilage volume and increased risk of cartilage defects [Teichtahl et al. 2008; Wang et al. 2007]. Over 2 years, fat mass and percentage of fat mass were associated with an increased rate of patella cartilage volume loss [Teichtahl et al. 2009b].

In a cross-sectional sample of 190 randomly selected subjects, serum levels of leptin were negatively associated with tibial cartilage volume, and the association between body mass index and cartilage volume disappeared after adjustment for leptin [Ding et al. 2008]. These findings suggest that reduced cartilage volume associated with obesity is related to leptin and thus is hormonally mediated [Ding et al. 2008]. A longitudinal study of 172 randomly selected subjects showed baseline interleukin-6 predicted loss of tibial cartilage volume, independently of TNF-α, and that change in interleukin-6 and TNF-α was associated with increased loss of tibial cartilage volume, suggesting that low-level inflammation plays a role in the pathogenesis of knee OA [Stannus et al. 2010].

Physical activity

Whether physical activity is beneficial or detrimental to articular structures of weight-bearing joints is unclear. Although epidemiological studies on radiographic OA have yielded conflicting evidence, recent studies using MRI to assess articular cartilage have shown a beneficial effect of physical activity on joint health.

A beneficial effect of exercise was seen in adults at risk of OA who completed an exercise program three times a week over 4 months, showing increased glycosaminoglycan content in their knee cartilage using dGEMRIC [Roos and Dahlberg, 2005]. Increased cartilage volume was present and accrued over 2 years by children aged 9–18 years who participated in more vigorous sports [Jones et al. 2000, 2003]. Similarly, in healthy adults without clinical knee OA, vigorous activity is associated with increased cartilage volume and reduced prevalence of cartilage defects [Racunica et al. 2007], greater muscle strength and fitness endurance is associated with reduced cartilage volume loss [Foley et al. 2007]. Conversely, and providing further support for the benefits of physical activity on cartilage properties, adults who became quadriplegic have shown more rapid loss of knee cartilage thickness over 12 months compared to age- and gender-matched controls [Vanwanseele et al. 2002, 2003]. These data suggest that physical activity is required for cartilage development and maintenance at the knee.

In terms of longer-term exposure to physical activity, participation in vigorous physical activity has been shown to be associated with a reduced rate of patella cartilage volume loss in those with no significant patella cartilage defects at baseline, suggesting that vigorous physical activity is beneficial to patellofemoral joints for people without pre-existing cartilage damage [Teichtahl et al. 2009a].

Smoking

More recently, smoking has been shown to be a risk factor for structural progression in OA [Amin et al. 2007; Davies-Tuck et al. 2009; Ding et al. 2007a], with increased cartilage volume loss and cartilage defects in knee OA, particularly in current smokers [Amin et al. 2007; Ding et al. 2007a]. Smoking (current and former) also predicted increased cartilage volume loss over 2 years in a population without knee OA [Davies-Tuck et al. 2009]. The mechanism by which smoking affects structures in knee OA is unclear, although components of tobacco smoke affect chondrocyte function by inhibiting cell proliferation and extracellular matrix synthesis in the spine [Akmal et al. 2004; Uematsu et al. 2001]. Similar mechanisms may operate in the knee. The role of smoking in OA pathogenesis is complicated by interactions with other factors including genetics [Ding et al. 2007a] and bone marrow lesions [Davies-Tuck et al. 2009]. A gene–environment interaction was observed in one study where smoking influenced cartilage volume loss and cartilage defects primarily in the adult offspring of parents who had knee replacements for primary knee OA [Ding et al. 2007a]. Nevertheless, the mechanism by which smoking affects OA pathology warrants further investigation.

Vitamin D

Vitamin D affects cartilage by stimulating proteoglycan synthesis and has been shown to be beneficial to joint health [Boyan et al. 1997]. A MRI study of older adults showed baseline levels and change in serum levels of vitamin D were both positively associated with tibial cartilage volume, and vitamin D insufficiency predicted cartilage volume loss over 2.9 years [Ding et al. 2009].

Role of MRI cartilage parameters in assessing treatment response in OA

Recently it has been shown that MRI can be used as an outcome measure in intervention studies of OA, monitoring the response to treatment. A multicentre clinical trial comparing the effects of licofelone and naproxen on the progression of OA found that although cartilage volume loss was observed in both groups over time, the loss was significantly smaller in the group receiving licofelone [Raynauld et al. 2009]. In contrast, there was no significant change in joint space narrowing over that time [Raynauld et al. 2009]. Cartilage mean thickness loss was also significantly less in the licofelone compared with the naproxen group at 12 and 24 months [Raynauld et al. 2008]. In a multicentre, randomized, double-blind, controlled trial of 69 patients with knee OA and clinical signs of synovitis, patients receiving chondroitin sulphate had significantly less cartilage volume loss than the placebo group as early as 6 months for the global knee, lateral compartment, and tibial plateaus, with significance persisting at 12 months [Wildi et al. 2011]. In a randomized, double-blind, placebo-controlled clinical trial of 117 patients with knee OA, Wluka and colleagues found no effect of supplementary vitamin E on tibial cartilage volume loss over 2 years [Wluka et al. 2002a]. An open-label pilot study examined the potential of celecoxib compared with a modelized historical control cohort for disease-modifying effect in the treatment of knee OA [Raynauld et al. 2010]. Over 12 months, no effect of celecoxib on medial and lateral tibial cartilage volume loss was observed [Raynauld et al. 2010]. A recent randomized-controlled clinical trial found no effect of lateral wedge insoles worn for 12 months on medial tibial cartilage volume loss over the same time period [Bennell et al. 2011].

A recent study examining the effect of intra- articular hyaluronic acid injections on articular cartilage assessed by MRI detected significant difference in patellofemoral cartilage defects over 8 weeks in the hyaluronic acid group, but it was not significantly different compared with the control group receiving intra-articular saline injections [Cubukcu et al. 2005]. However, the small sample size and short time span of the study limited its ability to draw strong conclusion on the effect of intra-articular hyaluronic acid injections on articular cartilage. A recently published single-blind clinical trial found that patients with symptomatic knee OA receiving 6 monthly intra-articular injections of Hylan G-F 20 showed reduced cartilage volume loss and reduced progression of cartilage defects over 2 years when compared with those receiving usual care for knee OA [Wang et al. 2011].

A randomized, placebo-controlled, double-blind pilot trial of collagen hydrolysate for mild knee OA examined the short-term changes in knee hyaline cartilage assessed by dGEMRIC. This study found increased dGEMRIC score in the medial and lateral tibial regions of interest in participants assigned to collagen hydrolysate but decreased dGEMRIC score in the placebo arm, with the changes between the two groups at 24 weeks reaching significance, suggesting that the dGEMRIC technique may be able to detect change in proteoglycan content in knee cartilage within 24 weeks [McAlindon et al. 2011]. These studies demonstrated that the effect of intervention could be established earlier using MRI cartilage measures compared with radiology in clinical trials.

The use of MRI has the potential to optimize surgical management of knee pathology. A recent study of knees with anterior cruciate ligament injuries found significant differences in cartilage morphology (cartilage volume, cartilage thickness and cartilage surface area) in surgically treated knees compared with those without surgery [Frobell et al. 2009]. An anterior cruciate ligament reconstruction performed within a mean of 6 weeks from injury was associated with increased cartilage volume and cartilage thickness in the central medial femur and decreased cartilage surface area in the trochlea femur at 12 months, compared with knees treated without reconstruction [Frobell et al. 2009]. By using MRI to monitor cartilage morphologies postsurgery, the finding of the study suggest a delayed structural restitution in anterior cruciate ligament reconstructed knees, and that the timing of surgery may be important in minimizing adverse structural changes postsurgery. The study also suggests a possibility that surgically treated patients may require longer recovery period prior to rehabilitation since there is ongoing structural remodelling in the traumatic knee.

Conclusion

Although the focus of this review is on the role of MRI in assessing articular cartilage, it is clear that the real strength of MRI has been in its ability to assess the entire joint, including cartilage, subchondral bone, menisci, synovial tissue, and ligaments. OA has been recognized as a disease of the whole joint. Emerging evidence has suggested that subchondral bone expansion and lesions (bone marrow lesions, bone cysts, bone attrition) and meniscal pathologies (extrusion and tear) may predict cartilage loss and play a role in the pathogenesis of OA [Crema et al. 2010; Davies-Tuck et al. 2010; Ding et al. 2007b, 2007c; Hunter et al. 2006; Neogi et al. 2009; Tanamas et al. 2010]. Using radiographs to assess OA, a lot has been learned about the epidemiology of OA. However, it is clear that simply using radiographs is limited in assessing cartilage change as it is unable to detect early cartilage changes and it is insensitive to change. The use of MRI has allowed us to extend what has already been learned about the epidemiology of OA. Superior to radiographs, MRI modality allows accurate visualization of all joint components and is sensitive to longitudinal changes. Thus, we are able to determine the associations between modifiable factors and joint structures at an early stage and investigate the interrelationships among different joint structures, which have the potential for better understanding of the pathogenetic mechanisms in OA.

MRI has the potential to detect early cartilage changes, in particular, the changes in cartilage matrix composition at stages when cartilage damage is still reversible and treatable, although further research is needed to determine the validity, reproducibility, feasibility and clinical significance of these biomarkers in the assessment of early cartilage damage. MRI enables quantitative assessment of articular cartilage noninvasively, thus representing a useful approach to determining the risk factors for OA, identifying the potential preventive strategies, monitoring the natural history of OA and evaluating the effect of therapeutic agents.