Magnetic Resonance Imaging of Trabecular Bone Structure
2002; Lippincott Williams & Wilkins; Volume: 13; Issue: 5 Linguagem: Inglês
10.1097/00002142-200210000-00004
ISSN1536-1004
Autores Tópico(s)Bone and Joint Diseases
ResumoINTRODUCTION Osteoporosis is a metabolic disorder characterized by a loss of bone mineral and characterized by the occurrence of atraumatic fractures of the vertebrae, fall-related hip fractures, and Colles fractures of the distal radius. Although the skeleton is composed of about 80% cortical bone, trabecular bone is highly responsive to metabolic stimuli and has a turnover rate approximately eight times that of cortical bone (1), thus making it a prime site for detecting early bone loss and monitoring response to therapeutic interventions. The primary imaging modalities used to measure bone properties and assess osteoporotic status, especially in the clinical arena, are x-ray or ultrasound based. In x-ray–based imaging techniques, the image intensity reflects tissue density; in ultrasound images, the image intensity is a depiction of the attenuation and speed alteration of sound waves. Both these methodologies in their clinical utilization reflect the density of bone mineral, and often the contribution of cortical and trabecular bone density cannot be determined. In the context of bone strength, the three-dimensional (3D) architecture of the bone is of importance. In the realm of macrostructures, the shape and the cross-sectional area of the bone itself contribute to bone strength; for example, in the femur the area of the femoral neck has some bearing on fracture occurrence. At the microarchitectural level, the thickness of the cortical bone, the 3D distribution of trabecular bone, the complexity, and connectedness of the trabecular network are parameters of importance. These features, coupled with measures of crystalline structure, bone matrix composition, and bone turnover, have been loosely termed "bone quality." It has become widely accepted that bone quality, which depends on several factors pertaining to both cortical and trabecular bone, may be important in the study of osteoporosis, both in predicting fracture risk and assessing therapy. The 3D, noninvasive imaging capabilities of magnetic resonance imaging (MRI) have been widely used clinically to assess and diagnose osteoporotic and vertebral fractures. Figure 1 shows an example of clinical images depicting the morphology and signal differences that are seen on MR images of vertebral fractures. In recent years, MRI has been developed to assess the characteristics of trabecular bone. It permits not only the depiction but also the quantification of trabecular bone structure, and hence its biomechanical properties. MR can be used to assess the properties of trabecular bone in two different ways. The first is an indirect measure, often termed relaxometry or quantitative magnetic resonance (QMR). This method takes advantage of the fact that trabecular bone alters the adjoining marrow relaxation properties in proportion to its density and structure and thereby provides information regarding trabecular bone network. The second is the direct visualization of the dark, trabecular bone, which, because of its low water content and short MR relaxation times, appears in stark contrast to the bright marrow fat and water in high-resolution MR images.FIG. 1.: T2-weighted MR image obtained through the spine of a 84-year-old female subject with an atraumatic fracture of L4 vertebra. (Courtesy of Cynthia Chin, Department of Radiology, University of California, San Francisco, California [UCSF]).QUANTITATIVE MAGNETIC RESONANCE Cortical bone and trabecular bone have short intrinsic T2 (relaxation time) values, low water content, and thus relatively low MR-detectable magnetization. In MR, the presence of bone in proximity to bone marrow results in a modification of the marrow relaxation times T1 and T2. The magnitude of T1 modification depends on the surface area-to-volume ratio of this bone-marrow interface, increases at higher magnetic field strengths, and increases when the number of bone and marrow interfaces increase, that is, as bone density increases. The magnitude of T2 relaxation time changes are governed by similar mechanisms as T1 relaxation but also are affected by processes such as diffusion. Magnetic susceptibility of trabecular bone is substantially different from that of bone marrow. This gives rise to localized inhomogeneities in the magnetic field, which depends on the number of trabecular bone-marrow interfaces, the size of the individual trabeculae, and the field strength. The diffusion of water in these magnetic field inhomogeneities results in an irreversible loss of magnetization and thus shortens the marrow relaxation time T2. This effect also depends on magnetic field strength and is greater at higher magnetic fields. In addition to these effects on marrow relaxation there is an additional effect that may occur in the presence of trabeculae, that is, the modification of the marrow relaxation time T2*. In specific types of MRI sequences, gradient-echo sequences, for example, in addition to diffusion-mediated loss of magnetization, the magnetization is further lost irreversibly as a result of the field inhomogeneities. This results in a characteristic relaxation time T2*, which includes the additional contribution due to field inhomogeneities as well as the T2 relaxation properties. This effect forms the basis of QMR. IN VITRO QMR STUDIES The impact of bone on the MR properties of marrow was first investigated in an in vitro experiment at a field strength of 5.8 T by Davis et al. in 1986 (2). Bone from autopsy specimens was ground up and sifted into a series of powders with graded densities ranging from 0.3 to 0.8 g/cc and then immersed in normalized saline or cottonseed oil to simulate bone marrow. Spectroscopic techniques showed that as the density of suspended bone increased, T1 relaxation time decreased as a result of an increase in surface area-to-volume ratio (increased surface interactions increased because higher bone density was comprised of smaller particles, thus more surface area). As the suspended bone density increased, there were concomitant increases in the magnetic field inhomogeneities and thus decreases in T2*, an effect more profound than on T1. At a field strength of 0.6 T, Rosenthal et al. (3) showed that in excised cadaveric specimens the relaxation time T2* of saline present in the marrow spaces was shorter than that of pure saline. Calibration of T2* with measures of bone mineral density (BMD) have been undertaken both in vitro (3–9) and in vivo (10–14). Dried excised specimens of vertebral bodies have been immersed in saline, an emulsion of peanut oil, and water to replicate marrow (4). To measure susceptibility-mediated effects on marrow relaxation times T1, T2, and T2*, experiments have been conducted at 1.5 T. Standard single-energy quantitative computed tomography (QCT) measures of BMD were then correlated with measures of T1, T2, and T2*. Whereas at 1.5 T there was no variation of the estimated T1 or T2 with trabecular density, at a lower field of 0.1 T Remy and Guilot (15) demonstrated a dependence of T2 on BMD. The investigators also found that 1/T2* increased with bone density at a rate of 0.20 ± 0.02 s−1/mg/cc (correlation coefficient 0.92, p < 0.0001) (4). Several investigators further correlated T2* in the both animal and human vertebrae (6–8) and ovariectomized rat models (9) and found correlations with measures of BMD and trabecular separation. They also found a dependence on the orientation of the trabeculae in the magnetic field. T2* variations with bone density are dependent on the spatial resolution at which the images are obtained (10), as well as on the 3D distribution of the trabecular bone, or structure, as shown in computer studies (5,16) and phantom experiments (3,17–19). Furthermore, the choice of echo time, the precise model selected to obtain the T2* relaxation time measure, affects the rate of change of T2* with bone density (19,20). In the area of osteoporosis, the biomechanical properties of trabecular bone are of ultimate importance. Using specimens from the human tibia (21) and vertebrae (22), it has been shown that 1/T2* increases linearly as the elastic modulus increases. However, the elastic modulus increases at a rate of 19.9 MPa/s−1 (22) in the vertebral specimens but at a rate of only 8.7 MPa/s−1 in the tibial specimens (21). However, this is not surprising because correlation between biomechanical strength and bone density in vertebral specimens and tibial specimens have shown differences, possibly due to regional variations and heterogeneity of tibial bone and differences in trabecular orientation and structure (23). Correlations between ultimate compressive strength and T2* have been studied in porcine bone (6) and human vertebral samples (24). These studies demonstrate strong correlations between ultimate strength and measures of relaxation time. All of these studies are indicative of the potential use of T2* as an indirect measure of bone strength, a parameter of importance in osteoporosis. IN VIVO QMR STUDIES Clinically, Sebag and Moore(25) showed qualitatively that bone marrow in the presence of trabecular bone showed increased signal loss in gradient-echo images, where T2* effects predominate. Subsequently, quantitative estimates of T2* in regions of varying bone density, such as in the epiphysis, metaphysis, and diaphysis, were measured by Ford et al. (12) using a technique known as interferometry, and localized proton spectroscopy. They demonstrated that in the distal femur of a normal volunteer, T2* in the epiphysis was 10.5 milliseconds, compared with 14.3, 72.2, and 79.9 milliseconds respectively, in the metaphysis, diaphysis, and subcutaneous fat. In a small sample size, it was shown that T2* values may distinguish osteoporotics from normals (11). In vivo calibration of T2* with trabecular bone density has been obtained from coincident measurements in the forearm, distal femur, and proximal tibia using MR and QCT (26). The relaxation rate, 1/T2*, increased as the bone density increased, at a rate 0.20 ± 0.01 s−1/mg/cc (correlation coefficient 0.88, p < 0.0001). In a study of six normal and nine postoophorectomy patients, Sugimoto et al. (13) showed a change in 1/T2* with bone density at a rate of 0.114 s−1/mg/cc. Fransson et al. (14) correlated T2* in the tibia with measures of BMD in the proximal femur and calcaneal ultrasound. The investigators found good correlations between T2* with BMD, but relatively lower correlations with ultrasound measures. This could be due to the significant heterogeneity of bone structure in the calcaneus, as well as in the tibia, and the fact that the ultrasound measure is a single-point measure and could be measuring a small and variable region between subjects. The heterogeneity in the bone density and its impact on T2* in the calcaneus was quantified in vivo by Guglielmi et al. (27), who showed that the shortest relaxation time occurs in the superior talar region (Fig. 2A), corresponding to the highest BMD. They also demonstrated a linear correlation between MR and dual x-ray absorptiometry (DXA) measurements (r = .66 for 1/T2* versus BMD). Song et al. (28) corroborated the regional differences of T2* in the calcaneus and showed the impact of trabecular bone orientation on marrow relaxation time. In a case control study, T2* measures of the proximal femur distinguished between subjects with hip fractures and normal subjects (29), as shown in Figure 2B. In addition, a combination of relaxation time measures and BMD improved the ability to discriminate subjects with vertebral fractures from those without (30). Use of relaxation time measures as an indirect means of characterizing trabecular bone and to some extent trabecular bone orientation has evolved and grown in the last several years, and it has potential use in the study of osteoporosis.FIG. 2.: A: Coronal image through the proximal femur showing the T2* values in different regions in different groups of subjcts (premenopausal, postmenopausal normal, and postmenopausal fracture subjects) showing significant differences between premenopausal and postmenopausal women, and between postmenopausal normal and hip fracture patients. [Adapted from Link et al. (29)]. A, Neck: 14.7 ± 1.0 ms (Pre), 18.5 ± 3.0 ms (Post)*, 21.0 ± 3.3 ms (Post, Hip Fx)**, significant difference between *Pre and Post and **Post and Post, Hip fx;B, Wards: 15.4 ± 1.6 ms (Pre), 18.7 ± 2.4 ms (Post)*, 22.7 ± 3.4 ms (Post, Hip Fx)**, significant difference between *Pre and Post and **Post and Post, Hip fx;C, Trochanter: 14.7 ± 1.0 ms (Pre), 18.6 ± 2.5 ms (Post)*, 19.8 ± 3.7 ms (Post, Hip Fx), significant difference between *Pre and Post. B: Sagittal image through the calcaneus showing the heterogeneity of T2* values in different regions. [Adapted from Guglielmi et al. (27)]. A, Superior Region: 9.52 ± 0.79 ms;B, Posterior Region: 11.58 ± 2.42 ms;C, Anterior Region: 11.75 ± 1.26 ms;D, Inferior Region: 12.25 ± 1.35 ms.Other methods of characterizing marrow relaxation properties and their impact on discerning trabecular bone properties have been studied. These methods rely on further analyzing the MR signal and, rather than calculating a single relaxation parameter, quantifying either a distribution of relaxation times or analyzing in detail the phase of the MR signal. Using the difference in phases in the MR images, Allein et al. (31–34) showed another potential method for measuring the impact of trabecular bone density and structure on marrow properties. Fantazzini et al. (35) calculated a distribution of relaxation times, in an in vitro setting, and ascribed differences to relaxation time distributions to differences in trabecular bone structure measures. HIGH-RESOLUTION IMAGING OF TRABECULAR BONE The marrow surrounding the trabecular bone network, if imaged at high resolution, reveals the trabecular network as seen in the representative image (Fig. 3A). In the figure, which is displayed in inverse gray scale, the bright network represents the trabecular bone network, whereas the lower-intensity background represents marrow-equivalent material in the trabecular spaces. Using such images, multiple different image processing and image analysis algorithms have been developed. The goal of all of these methods was to quantify the trabecular bone structure in two or three dimensions. Many measures have been derived so far; some of them are synonymous with the histomorphometric measures such trabecular bone volume fraction (BV/TV), trabecular thickness (Tb.Th), trabecular spacing (Tb.Sp), and trabecular number (Tb.N). Others include connectivity or Euler number, fractal dimension, tubularity, and maximal entropy. Some of the results obtained using these descriptors, and their utility thus far, will be summarized later. Another aspect of computerized analysis deals with visualization and depiction of the 3D structure imaged using MR. There are numerous ways of surface and volume rendering such images. Figure 3B shows an example of 3D renderings of an MR image obtained at 1.5 T using a clinical scanner. The specimen was obtained from an elderly subject. Larger trabecular spaces and loss of trabecular bone in the elderly subject are clearly visible.FIG. 3.: A: Image obtained at resolutions of 117 × 117 × 300 μm at 1.5 T of a vertebral body sample with the structural parameters, bone mineral density, and elastic modulus measured along two axes as shown in the figure. B: Vertebral body images obtained from samples from a 69-year-old man were surface rendered. The loss of trabecular bone volume in the elderly subject (trabecular bone volume = 14%) is apparent by the large spaces seen in the sample on the right. (Courtesy of Olivier Beuf, Andres Laib, Magnetic Resonance Science Center, UCSF.).A number of calibration and validation studies (in vitro and in vivo) have been undertaken in which MR-derived measures of structure are compared with measures derived from other modalities, such as histology, micro-CT, BMD, and biomechanics. One of the primary issues in MR-derived visualization and quantitation of structure arises from the fact that the spatial resolution of the MR images often is comparable to the thickness of the trabecular bone itself, which gives rise to partial volume effects in the image. Thus, the image may not depict very thin trabeculae or may represent an average or a projection of a few trabeculae. Recognizing that MR-derived measures are not identical to histologic dimensions, a major focus in the field has been using established measures to investigate the resolution dependence of MR-based measures and then calibrating MR-derived measures of bone structure. IN VITRO HIGH-RESOLUTION IMAGING OF TRABECULAR BONE Hipp et al. (36) compared the morphologic analysis of 16 bovine trabecular bone parallelepipeds using both 3D MR reconstruction (92 × 92 × 92 μm3) and 2D optical images (23 × 23 μm2) of the six faces of the samples. The volume fraction measured by MRI was linearly related to and not different from the average area fraction measured from optical images (R2 = 0.81, p = 0.96). Resolution-dependent effects in comparing structure measurements of human distal radius specimens with MRI of 156 × 156 × 300 μm3 and high-resolution x-ray tomographic microscopy (XTM) at resolutions of 18 × 18 × 18 μm3 (37) have been undertaken. In these studies, high-resolution XTM images were used to simulate resolution degradation using low-pass filtering. It was observed that a decrease of the spatial resolution induced overestimation of BV/TV and Tb.Th, and underestimation of Tb.Sp and Tb.N. Using a set of 18 trabecular bone samples, structure measures derived from MRI (117 × 117 × μm3) and optical imaging (20 × 20 μm2) were compared (38). It was shown that there was good correlation between MR-derived and optical imaging-derived measures such as Tb.Sp (r = 0.89, p < 0.01) and Tb.N (r = 0.78, p < 0.01), moderate correlation for BV/TV (r = 0.69, p < 0.01), and very poor correlation for Tb.Th (r = 0.06, p = 0.84). These correlations indicate that MR images can depict trabecular bone structure, although the absolute measures differ from the measures obtained at higher resolution, except in the measures of trabecular thickness. This is explained by the fact that the image resolution is comparable to the dimensions of the trabeculae being measured, and a small sampling error, or a partial volume effect, in the estimation leads to a large percentage or fractional error. The effect of slice thickness on standard morphologic measurements was investigated by Kothari et al. (39), who artificially degraded from 100 to 1000 μm (with a spatial resolution of 100 × 100 μm2) using 3D optical reconstruction of trabecular bone specimens with initial isotropic resolution of 40 × 40 × 40 μm3. Measurements such as BV/TV, Tb.Sp, and Tb.N had weak resolution dependency, whereas Tb.Th was strongly affected by the resolution rising rapidly with increasing slice thickness, requiring very high resolution for precise evaluation. Vieth et al. (40) compared standard morphologic measurements (BV/TV, Tb.Th, Tb.Sp, Tb.N) of 30 calcaneus specimens using MRI (195 × 195 μm2 in-plane resolution and 300/900 μm slice thickness) and contact radiographs (digitized with 50 × 50 μm2 spatial resolution) of sections obtained from the same specimens. The results of this study show that MR-based measurements were significantly correlated with those obtained from digitized contact radiographs. However, partial volume effects due to slice thickness as well as image postprocessing (thresholding) had substantial impact on these correlations; the thicker the slice, the poorer the correlation. Lin et al. (41) confirmed correlation between structure parameters derived from MR images and serial grinding images (three specimens) and established that the heterogeneity of calcaneal bone structure, as determined from MR images, is real and correlated to the magnitude of the spatial heterogeneity using higher resolution microscopic images. The accuracy of a new model-independent morphologic measure, based on the distance transformation technique applied to high-resolution MR images of human radius specimens with in vivo resolution of 156 × 156 μm2 in plane and 300 or 500 μm in slice thickness, were investigated by Laib et al. (42). These measures were compared with high-resolution micro-CT images (34 × 34 × 34 μm3), and good correlations were found between the two sets of measurements, with the best R2 = 0.91 for Tb.N. This study showed that new model-independent morphologic measures applied to high-resolution MR images give better accuracy than standard morphologic measurements. From the same set of radius specimens, Pothuaud et al. (43) showed the ability of 3D-line skeleton graph analysis to characterize high-resolution MR images obtained with 156 × 156 × 300 μm3 resolution. Using a clinical scanner, Majumdar et al. (37) derived structure measurements from MR reconstructions (156 × 156 × 300 μm3) of human distal radius cubes and evaluated the explanatory power of these measurements in the prediction of the elastic modulus evaluated from nondestructive testing of the same specimens. Although structure characteristics were not accurately restored (compared to 18 × 18 × 18 μm3 XTM-based measures), significant correlations were established with the mean elastic modulus (evaluated in the three directions of the specimens): r = 0.96 (p = 0.002) for BV/TV, r = 0.82 (p = 0.05) for Tb.Th, r = 0.95 (p = 0.003) for Tb.Sp, and r = 0.78 (p = 0.06) for Tb.Th. The feasibility of using MRI at a resolution of 117 × 117 × 300 μm3 to quantify structure, as well as to better predict mechanical properties, was established by Majumdar et al. (38) using a set of 94 specimens of several skeletal sites, with a wide range of bone densities and structures. Among several results reported in this study, it was shown that MR-based structural measures, used in conjunction with BMD (evaluated from QCT measures), enhanced prediction of bone strength. Using a stepwise regression model, including structural parameters in addition to BMD, resulted in improvement of the prediction of the mean elastic modulus (evaluated from nondestructive testing). The adjusted correlation coefficient increased from 0.66 to 0.76 for all the specimens, 0.71 to 0.82 for vertebral specimens, and 0.64 to 0.76 for femoral specimens. MR images (117 × 117 × 300 μm3) of vertebral midsagittal sections of lumbar vertebrae and standard morphologic parameters were calculated by Beuf et al. (24). Ultimate stress was estimated in two perpendicular directions (horizontal/vertical) using compression testing applied to two cylindrical samples drilled in each vertebra close to the midsagittal section. All the morphologic parameters were correlated to both horizontal and vertical ultimate stresses (r > 0.8, p < 0.001). Using the same set of MR images, Pothuaud et al. (44) compared topologic parameters evaluated from the newly developed technique 3D-LSGA: line skeleton graph analysis, with elastic moduli evaluated from finite element analysis (FEA). High correlations were found between topologic parameters and elastic moduli (R2 = 0.94–0.95), as well as between topologic parameters and ultimate stress (R2 = 0.85 for horizontal direction, R2 = 0.76 for vertical direction). Link et al. (45) used texture parameters measures on high-resolution MR images (156 × 156 × 300 μm3) of proximal femur and spine specimens. Whereas the correlation between YM and BMD was R2 = 0.66 for the spine specimens and R2 = 0.61 for the femur specimens, a multivariate regression model combining both BMD and texture parameters increased the correlation to R2 = 0.83 for spine and to R2 = 0.72 for femur. Although this study shows that gray-level texture evaluation could be useful for high-resolution MR applications, the relationships between the gray-level characteristics and the true structure needs to be explored. ANIMAL STUDIES USING HIGH-RESOLUTION IMAGING OF TRABECULAR BONE Several investigators conducted studies in in vitro animal models of osteoporosis and in vivo animal studies. White et al. (46) showed that MRI (on a 2.0-T scanner) could assess structure in rats both in vitro and in vivo. Luo et al. (47) used in vivo MRI to monitor time-dependent changes of structure in rat tibia induced by glucocorticoid therapy. MRI was performed on a 4.7-T scanner with a spatial resolution of 51 × 51 μm2 in an axial plane, 100 μm along the tibia. Significant decrease in BV/TV and Tb.N, and increase in Tb.Sp were observed after 15 days in the treated group compared with the control group. These changes reversed between 15 and 30 days, which was consistent with previous histomorphometry studies. Takahashi et al. (48) have used μ-MRI at 9.4 T to monitor the effects of preventive agents in an ovariectomized rat. The results were in accordance with literature data obtained in similar models by histomorphometry, proving that μ-MRI allows the characterization of trabecular bone structure in small animals with sufficient accuracy. Using a small-bore scanner, Borah et al. (49) obtained MR images at an isotropic resolution of 85 μm to analyze the 3D trabecular bone structure and mechanical properties of vertebral specimens of young and mature mini-pigs. Accurate characterization of the 3D structure based on newly developed nonmorphologic parameters as well as FEA have shown significant differences between the two groups of mini-pigs. Furthermore, apparent Young modulus, evaluated from experimental test and FEA modeling, were similar and structure parameters contributed to the prediction of compressive strength evaluated experimentally. IN VIVO HUMAN STUDIES USING HIGH-RESOLUTION IMAGING OF TRABECULAR BONE In vivo, the skeletal sites most commonly imaged are the radius (37,50–55) and calcaneus (40,56–59). The distal radius is a site with a large quantity of trabecular bone and a common site for osteoporotic fractures. It is easily accessible with localized detection coils and subjects are able to comfortably tolerate immobilization for the period required for high-resolution imaging. Calcaneus, although not a typical site for osteoporotic fractures, has been used with success to predict fracture at other sites, and this skeletal site is well adapted to high-resolution MRI. The phalanges recently have been of increased interest as a site for bone density measurement (60,61) and can be imaged by high-resolution MRI. Figure 4A shows representative axial scans through the distal radius;Figure 4B shows a representative scan through the calcaneus. The distal femur and distal and proximal tibia also contain a large amount of trabecular bone. Changes in gait, onset of arthritis (62), and systemic changes in bone density and quality also may be depicted in these regions.FIG. 4.: A: High-resolution MR images obtained in the axial plane through the distal radius, at 1.5 T. The images are obtained for two subjects and show sections from the distal radius starting closer to the joint line and into the shaft. The spatial resolution of the images is 156 × 156 × 500 μm. (Courtesy of David Newitt, Department of Radiology, UCSF.). B: High-resolution MR image obtained in the sagittal plane of the calcaneus, at 1.5 T. The spatial resolution of the images is 195 × 195 × 500 μm.The image contrast can be manipulated on MR images based on the specific pulse sequence used, and the appearance of trabecular bone can be varied based on whether a spin-echo or gradient-echo sequence is used (63). The high susceptibility difference between bone and marrow induces susceptibility artifacts at their boundary, which in the case of in vivo imaging could have a high impact on the bone structure quantification (63). Although spin-echo images may be preferable to reduce this effect, gradient-echo images acquire an equivalent volume in considerably less time and can be exploited in vivo at several skeletal sites. By optimizing the pulse sequence timing (short echo time) in gradient-echo imaging, one can attempt to minimize the susceptibility artifact. Although there may be predictable improvements in the field strength of MR scanners, pulse sequence design, and radiofrequency coil optimization, the resolution achievable in vivo on clinical scanners is limited by the inherent signal-to-noise ratio of the images. Quantitative evaluation of structure from these images also constitutes a major area of research. Processing of high-resolution MR images generally consists of several stages (53) and may include some or all of the following: (1) preprocessing, (2) segmentation to define the region of trabecular bone, (3) texture analysis or direct gray-level analysis, (4) binarization into bone and marrow phases, and (5) derivation of structure parameters. Other stages, such as a manual region adjustment for reproducibility studies or image registration for longitudinal studies, also may be necessary. In the case of high-resolution MR images of structure, Newitt et al. (53) showed that each stage needs to be standardized and normalized to ensure a high degree of reproducibility. In particular, these authors describe a standardized analysis system with considerable reduction of human interaction. The efficiency of this system was evaluated in terms of reproducibility (2–4%) and successfully applied in several cross-sectional (52,59,64–66) and longitudinal studies (67,68). Some noise reduction-based preprocessing techniques have been applied before the binarization stage, such as low-pass filtering (50) or histogram deconvolution (69). In addition, use of some postprocessing schemes after binarization, based on either morphologic criterion relative to the shape and morphology of the trabeculae (70) or on topologic criterion relative to the numbers of bone and marrow components (71), were applied. Wu et al. (72) proposed a sophisticated histogram model taking into account the partial volume effect characterizing MR images, using a probabilistic approach. The relative solid and void (or marrow) fractions of each voxel were evaluated, allowing classification of each voxel independently, as being either in the solid phase or in the void (or marrow) phase. Hwang et al. (73) used spatial correlation analysis, which is based on the probability of finding bone at a specific location. Some morphologic parameters were deduced from this analysis, such as intertrabecular spacing, contiguity, and tubularity. In a set of 20 subjects classified as normal or osteoporotic based on vertebral deformity, a combination of some of these parameters was predictive of the vertebral deformity (R = 0.78, p < 0.005) (74). Gray-level histogram-based binarization is commonly used with high-resolution MR images of trabecular bone, but because it is a scheme based on a single threshold value applied with a limited resolution (compared to the mean size of the trabeculae) it may lead to global thickening of trabeculae or loss of the thinner trabeculae. For example, if image resolution is 100 μm and trabecular dimensions are on the order of 100 μm, an error of one voxel may be reflected as a 100% error in the estimated trabecular width. Similarly, a trabecula that is 50 μm thick will be detected either as a 100-μm structure or as no trabecula (marrow phase). Recognizing that it is not possible to accurately reconstitute the "true" trabecular bone structure from high-resolution MR images, Majumdar et al. (37) introduced the notion of "apparent" trabecular bone network. Of note was the fact that while noting the "apparent" network is not identical to the "true" histologic structure, it nonetheless reflects some "apparent" morphologic and topologic properties that are highly correlated to the "true" one (37,39). More recently, distance transformation technique was applied to high-resolution in vivo MR images of the distal radius (156 × 156 × 500 μm3) in postmenopausal women (75). Morphology-based parameters were evaluated without assumption of any structure model, and the most significant parameter in distinguishing subjects with vertebral fracture (n = 88) from those without vertebral fracture (n = 60) was the intraindividual distribution of separation (standard deviation of the trabecular bone separation parameter). Using receiver operating curve analysis, the competence of this parameter was comparable to radius or spine BMD measures, but not as pertinent as the competence of hip BMD alone. FEA involves calculation of mechanical properties of structure by using finite element models, generally obtained by converting each voxel to equally shaped brick eight-node elements (76,77). Nevertheless, such voxel-based FE models consist of a large number of elements and special purpose FE solvers are required to evaluate mechanical properties with reasonable calculation conditions. Although FEA initially was developed and optimized for in vitro applications, recent advances in FEA techniques allow direct evaluation of mechanical properties from high-resolution in vivo MR images of trabecular bone (65). In a longitudinal study consisting of a group of postmenopausal women who were either normal or osteopenic based on their spinal or femoral BMD values and the World Health Organization criterion were studied at baseline and 1 year after randomized assignment to one of three treatment groups: placebo, 5 mg idoxifene (selective estrogen receptor modulator), or 10 mg idoxifene. The study was discontinued after the first year due to adverse effects of the drug; however, it provided the first set of longitudinal MR data on bone structure. In addition, BMD of the calcaneus, hip, and spine were measured using dual DXA (Hologic, MA, USA). Ultrasound measures speed of sound (SOS) and broadband ultrasound attenuation (BUA) were obtained (Sahara System, Hologic, MA, USA). Measures of bone turnover was assessed using urinary C telopeptide excretion, serum total osteocalcin, and serum type I collagen C-telopeptide. Axial MR images of the distal radius and sagittal MR images (500-μm slice thickness) of the calcaneus were used to derive measures of apparent trabecular bone volume fraction (App BV/TV), spacing, number, and thickness (App Tb.Sp, App Tb.N, App Tb.Th) and to build micro finite element models to determine the architecture-dependent directional Young modulus (E1>E2>E3) and the anisotropy ratios E1/E2, E1/E3, and E1/E2, as well as the shear moduli G12, G23, and G13. In the radius, using the baseline data Newitt et al. (54) showed in two groups of postmenopausal women (22 normal and 37 osteopenic) that the standard morphologic structure parameters as well as FEA-based Young and shear moduli were lower in the osteopenic group compared with the normal group, and the anisotropy of the Young moduli were higher in the osteopenic group compared with the normal group. Although FEA-based parameters were statistically correlated to bone volume fraction, a stepwise regression analysis showed that the combination of bone volume fraction with some of the morphologic parameters could improve the prediction of the mechanical properties. The baseline calcaneus data showed that the maximum elastic modulus (E1) in the calcaneus was in the medial-lateral direction (1.38 ± 0.6 GPa, average for all subjects). In the superior-inferior direction E2 was comparable in magnitude (1.2 ± 0.6 GPa), whereas in the anterior-posterior direction E3 was lower (0.4 ± 0.2 GPa). There was a strong correlation (r values between 0.93 and 0.82) between the elastic constants and App BV/TV. The association between E and App Tb.N and Tb.Th was positive but weaker, and, as expected, there was an inverse correlation with App Tb.Sp (r values between -0.31 and -0.26). Both E1/E3 and E2/E3 anisotropy ratios showed a negative association with BMD, BUA, and SOS, but a positive association with age and years since menopause. The longitudinal data are presented in Table 1. The percentage change in bone density, measures of bone turnover, and MR measures, as well as the FEA-derived elastic moduli, are shown. Due to its early termination, the study did not show any significance between the response of the placebo and treatment groups, although the trends of the impact of therapy are represented (Table 1).Comparing baseline and 1-year data in each group separately, significant differences were seen in some of the parameters. In fact, Van Rietbergen et al. (67) demonstrated such effects in longitudinal changes in mechanical parameters in the calcaneus.TABLE 1: Percentage changes (mean ± standard deviation) after 1 year treatment with SERM, IdoxifeneBeuf et. al. (62) studied trabecular bone structure in subjects with mild, moderate, or severe osteoarthritis of the knee joint and demonstrated significant differences between structure in the femur and tibia, which decreased with the degree of osteoarthritis. The authors found that the apparent BV/TV, Tb.N, and Tb.Sp in the femoral condyles could be used to differentiate healthy subjects or subjects with mild osteoarthritis from patients with severe osteoarthritis (p < 0.05). Among individuals, the authors demonstrated that the structural variation of the lateral and medial femoral condyle was indicative of the disease extent. CONCLUSION With the advent of higher-field magnets for clinical imaging and computerized image processing, MR promises to provide an important complement to standard methods of assessing osteoporosis and response to therapy. Over the last several years different magnetic resonance methods have emerged as a means of measuring trabecular bone structure and have been shown to play a role in the assessment of skeletal strength and integrity. Although in vitro studies in animal models are a mandatory part of technique development, the potential strength and role of MRI lies in its human in vivo applications. Not only is the use of MR an attractive alternative for assessing trabecular bone structure, but its potential for quantifying bone marrow composition (78–81) makes it an attractive modality for comprehensive characterization of age- or therapy-related metabolic changes in the skeleton. FIGUREFIG. 5.: High-resolution (195 × 195 × 500 μm) image through the human calcaneus. Dark striations reflect trabecular bone and bright signal is from bone marrow. The trabecular number is highest in the subtalar region coincident with the shortest T2 (Fig. 2b).
Referência(s)