MR pulse sequences for quantitative analysis of cartilage morphology
An MRI pulse sequence suitable for measuring cartilage morphology quantitatively must provide a high signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) for accurate delineation of the subchondral bone interface and articular surface, and no significant artefacts must be present. Measurements should be obtained at relatively short imaging times, in order to avoid motion artefacts and to be able to measure cartilage deformation directly after exercise (Tieschky et al. 1997). Because cartilage layers exhibit a mean thickness of only 1.3–2.5 mm throughout the human knee (Eckstein et al. 2001a; Hudelmaier et al. 2001) and even less so in other joints (Peterfy et al. 1995; Springer et al. 1998; Graichen et al. 2000, 2003; Al Ali et al. 2002), a high spatial resolution is required so that a sufficient number of image points (pixels) are available to characterize the thickness of the tissue throughout the joint surface, including areas with thin cartilage coverage. Increasing the resolution by a factor of two in three dimensions requires acquisition times to be increased by a factor of 64, if the SNR is to be kept constant. Although there is no current consensus on the optimal resolution for imaging cartilage morphology, a 1.5-mm section thickness and 0.3-mm in-plane resolution has been commonly used at a field strength of 1.5 T. The specific MR pulse sequences that has been most frequently employed for cartilage imaging is a T1-weighted spoiled gradient echo sequence [FLASH = fast low angle shot (Frahm et al. 1986) or SPGR = spoiled gradient recalled acquisition at steady state]. This sequence (Fig. 1a,b) is available on most clinical MRI systems and has been implemented either with frequency-selective spectral fat-suppression by a prepulse (Recht et al. 1993; Peterfy et al. 1994; Eckstein et al. 1996a; Cicuttini et al. 2000) or with frequency-selective water excitation (Hardy et al. 1998; Graichen et al. 2000; Burgkart et al. 2001; Glaser et al. 2001). Both techniques achieve effective fat-saturation, which is required to provide a sufficient dynamic range of the image contrast between the cartilage and its surrounding tissues, and to eliminate artefacts at the subchondral bone interface. New 3.0-T whole-body MR scanners now make it possible to perform quantitative cartilage imaging at higher field strength (Gold et al. 2004a,b; Eckstein et al. 2005a,b; Kornaat et al. 2005). A recent study has shown that measurements at 3.0 T are consistent with those at 1.5 T, and that the precision (reproducibility) of the measurements is slightly improved when exploiting the higher field strength to obtain a higher spatial resolution (1-mm slice thickness) at 3.0 T compared with 1.5 mm at 1.5 T (Eckstein et al. 2005a).
Figure 1. (a) Coronal MR imaging (slice thickness 1.5 mm, in-plane resolution 0.31 mm × 0.31 mm) acquired with a T1-weighted spoiled gradient echo sequence (FLASH = fast low angle shot; or SPGR = spoiled gradient recalled acquisition at steady state) with frequency-selective water excitation. (b) Segmentation showing the medial tibial cartilage in blue, the medial femoral condyle in yellow, the lateral tibia cartilage in green, and the lateral femoral cartilage in red. (c) Sagittal dGEMRIC image kindly provided by Dr Deborah Burstein, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, USA.
Download figure to PowerPoint
One of the great advantages of MRI (e.g. in comparison with histology) is that consecutive slices are contiguous and spatially aligned, so that three-dimensional (3D) parameters can be obtained that characterize cartilage morphology appropriately (Fig. 2). These parameters include cartilage volume, cartilage thickness (mean, maximum, standard deviation), cartilage surface area (or subchondral bone interface area) as a measure of bone size, cartilage surface curvature (joint incongruity) and others (Fig. 2). When reporting cartilage volume, one must keep in mind that this parameter depends on both the cartilage thickness and the cartilage surface area, and that only under conditions where the cartilage surface (or chondro-osseous interface area) is constant, do volume or thickness changes over time correspond. In cross-sectional studies, it is important to report cartilage thickness directly, or to normalize cartilage volume to the joint surface/bone interface area, in order to provide meaningful results. It has been shown, for instance, that gender differences in joint surface areas are substantially larger than those for cartilage thickness (Faber et al. 2001), a finding that is not evident from measuring cartilage volume alone. Also, it has been shown that cartilage thickness and cartilage surface areas are not closely associated in healthy individuals (Eckstein et al. 2001b); in other words, subjects in whom the articular cartilage occupies a larger surface area do not necessarily have thick cartilage and vice versa. Thus, one of these parameters cannot be estimated from the other – both must be measured as separate entities.
Figure 2. (a) Three-dimensional reconstruction of femoral and tibial cartilage from segmentations of contiguous MR images; (b) distribution pattern of cartilage thickness in the femur, determined independent of the original section orientation. The blue colour shows areas of thick cartilage, orange and red show areas of thin cartilage.
Download figure to PowerPoint
In order to derive quantitative data from a 3D, contiguous image set, an anatomical structure (the articular cartilage) must first be labelled, distinguishing it from its immediate relations (segmentation –Fig. 1b). Owing to the relatively low contrast in some areas of the joint surface (joint contact areas, the vicinity of synovial folds, tendons and ligaments, repair tissue, etc.), fully automated segmentation of cartilage is impractical from MR images. Various semi-automated image analysis techniques have been developed to date, each requiring different degrees of user interaction. Verification (and some degree of correction) by an experienced user is generally necessary on a section-by-section basis. The inability of current computer software reliably to identify structures in images that are evident to the experienced human eye may seem surprising. However, if one considers the great difficulties involved in automated speech recognition by computers, despite the tremendous efforts made by industry, one may appreciate the complexity of automated recognition in intricate image pattern identification. For these reasons, and because many slices must be acquired of one joint surface to obtain sufficient spatial resolution, cartilage segmentation is a time-consuming process currently requiring several hours of human interaction per knee data set.
After segmentation, computation of the cartilage volume is straightforward, by simple numerical integration of the number of voxels attributed to the cartilage during the segmentation process (Fig. 2). More sophisticated algorithms are then used to determine the cartilage thickness (Fig. 2) and joint surface area, which must account for out-of-plane deviations of these parameters. Computations should therefore be made in three dimensions, independent of the original section orientation. Extraction of cartilage surfaces also allows for the determination of geometric topography and curvature characteristics of diarthrodial joints (Ateshian et al. 1991). Mathematical descriptions of joint surfaces and articular cartilage layers can also be applied to derive computer models of human joints, by which the contact areas and surface stresses in joints may be estimated (Cohen et al. 1999, 2001) but these methods have not been yet applied to the study of the effects of exercise on cartilage and joint morphology.
Because these ‘global’ parameters (volume and mean thickness for an entire cartilage plate) may be relatively insensitive to regional/focal changes that affect only small portions of the surface, several investigators have presented techniques for displaying regional cartilage thickness patterns (Eckstein et al. 1995, 1996a,b, 1998a; Sittek et al. 1996; Cohen et al. 1999, 2003; McGibbon & Trahan, 2003) (Fig. 2). Changes over time (or differences between subjects) in regional cartilage thickness are, however, difficult to detect from subjective comparison of such thickness patterns, because only a limited number of thickness intervals can be displayed. In order to track local/regional thickness changes over time, registration techniques have therefore been proposed (Kshirsagar et al. 1998; Stammberger et al. 2000; Waterton et al. 2000; Lynch et al. 2001; Cohen et al. 2003; Raynauld et al. 2003). With these methods, the bone interface or other anatomical landmarks from two data sets are matched so that the thickness distribution can be compared on a point-by-point basis. Stammberger et al. (2000) reported a local mismatch of cartilage thickness for joint repositioning in the range of 0.5–1 mm. These local errors are relatively large in comparison with the absolute cartilage thickness in knee joint surfaces, but this is not surprising given that an anatomically complex structure is reconstructed and registered with data obtained from a limited number of sectional images.
Validation and reproducibility (precision) of quantitative analysis of cartilage morphology
The validity (accuracy) of qMRI of cartilage has been addressed in numerous studies over recent years and these have been carried out in unselected cadaver joints, amputated joints (Peterfy et al. 1994; Cicuttini et al. 1999) or knee joint of patients undergoing total knee arthroplasty (TKA) (Peterfy et al. 1994; Cicuttini et al. 1999; Burgkart et al. 2001; Graichen et al. 2004). TKA provides a unique opportunity for validating quantitative measurements, as patients can be imaged prior to surgery in vivo, and the tissue can be removed and analysed after the operation. Validation studies have been carried out in comparison with various reference methods, namely water displacement of surgically retrieved tissue (either direct or by employing Archimedes’ principle), anatomical sections obtained with high-precision band saws, computer tomography arthrography, A-mode ultrasound (not to be confused with clinical B-mode ultrasound), and stereophotogrammetry. Most of these comparative studies have reported close agreement between methods of measuring cartilage volume, with random errors (absolute pairwise over- or underestimation) vs. the respective reference method of about 5–10%. Validation studies have also been performed in other joints with thinner cartilage, such as the metacarpophalageal joint (Peterfy et al. 1995), the hip (McGibbon et al. 1998), the elbow (Graichen et al. 2000) and the shoulder (Graichen et al. 2003).
Precision errors are random errors that occur when repeated measurements of a parameter are taken under constant conditions. Highly reproducible techniques are required to resolve small changes (i.e. cartilage deformation) with statistical confidence. For qMRI of cartilage morphology, the precision depends on factors associated with image acquisition, and factors associated with image analysis. Differences in joint positioning are less critical than for projectional techniques (such as radiography), because the technique is 3D and the relevant quantitative measures are obtained from reconstructions of serial images rather than from projection onto one image plane. The lowest precision errors (CV%∼1%) have been observed for axial protocols of the patella (Eckstein et al. 2000b). Higher precision errors, by contrast, have been reported for analyses of the femoral condyles in sagittal scans (Eckstein et al. 2002b), whereas analysis of the total femur has usually been comparable with other joint surfaces of the knee. Precision errors of computations of the mean cartilage thickness throughout joint surfaces have been reported to be similar to those of cartilage volume (Stammberger et al. 1999; Hyhlik-Dürr et al. 2000; Burgkart et al. 2001; Eckstein et al. 2002b) as have those for quantification of cartilage surface areas (Hohe et al. 2002; Eckstein et al. 2002b).
MR protocols for compositional cartilage imaging
In addition to measuring cartilage morphology, there have been great efforts in using MRI to determine the composition of cartilage, namely the glycosaminoglycan (GAG) content, collagen content and orientation, and the interstitial water content. Attempts to determine the concentration of GAG include imaging of fixed charge density by using an intravenous injection of the charged clinical MRI contrast agent Gd(DTPA)2–. If Gd(DTPA)2– is allowed to penetrate into cartilage, a process that has been estimated to last for about 90 min after injection, it distributes inversely with the GAG concentration (Fig. 1c). Because full penetration is required, the technique has been termed delayed gadolinium enhanced MRI of cartilage (dGEMRIC) (Gray et al. 2004). When tissue is placed in a magnetic field, magnetic moments of the protons are aligned, resulting in a net magnetic moment. This equilibrium is then disturbed by transmitting another magnetic field at the same frequency as the rotations of the protons for a very short time. The return to equilibrium of the magnetic moments after this pulse is strongly affected by molecular interactions of the nuclei with their surroundings and can be exploited for imaging cartilage composition (Burstein & Gray, 2003). Two time constants are relevant in this context, the longitudinal (T1) and transverse relaxation time (T2). When probing T1 in cartilage in the presence of fully penetrated Gd(DTPA)2– (dGEMRIC), one can estimate the GAG content of the tissue (Fig. 1c). dGEMRIC has been validated in basic science and clinical studies through comparison with biochemical and histological measures of GAG (Bashir et al. 1997, 1999; Tiderius et al. 2003; Williams et al. 2004). Another technique that has been successfully explored is T2 mapping (Mosher & Dardzinski, 2004). T2 can be obtained without the presence of a contrast agent, but cannot be attributed to a single constituent of cartilage composition. T2 has been shown to provide a quantitative measure of cartilage interstitial fluid and its interaction with the solid components of the extracellular cartilage matrix, in particular with collagen content and orientation, whereas there is little to no sensitivity to changes in GAG concentration (Mosher & Dardzinski, 2004). Spatially resolved cartilage T2 maps have been shown to be correlated with the regional water content of the deep and mid zones of cartilage (Lusse et al. 2000) but recent data suggest that collagen fibre anisotropy is the dominant factor related to regional differences in T2. T2 of the superficial zone of cartilage was shown to change with aging (Mosher et al. 2004).