Quantitative high-resolution magnetic resonance imaging reveals structural implications of renal osteodystrophy on trabecular and cortical bone


  • Felix W. Wehrli PhD,

    Corresponding author
    1. Laboratory for Structural NMR Imaging, Department of Radiology, University of Pennsylvania Medical Center, Philadelphia, Pennsylvania
    • Department of Radiology, University of Pennsylvania Medical Center, 3400 Spruce St., 1 Founders Pavilion, Philadelphia, PA 19104
    Search for more papers by this author
  • Mary B. Leonard MD,

    1. Department of Pediatrics, Epidemiology and Biostatistics, University of Pennsylvania Medical Center, Philadelphia, Pennsylvania
    Search for more papers by this author
  • Punam K. Saha PhD,

    1. Laboratory for Structural NMR Imaging, Department of Radiology, University of Pennsylvania Medical Center, Philadelphia, Pennsylvania
    Search for more papers by this author
  • Bryon R. Gomberg PhD

    1. Laboratory for Structural NMR Imaging, Department of Radiology, University of Pennsylvania Medical Center, Philadelphia, Pennsylvania
    2. Department of Radiology, Hadassah Medical School, Jerusalem, Israel
    Current affiliation:
    1. Department of Radiology, Hadassah Medical School, Jerusalem, Israel
    Search for more papers by this author



To explore the potential role of micro-magnetic resonance imaging (μ-MRI) for quantifying trabecular and cortical bone structural parameters in renal osteodystrophy (ROD), a multifactorial disorder of bone metabolism, traditionally evaluated by bone biopsy.

Materials and Methods

Seventeen hemodialysis patients (average PTH level = 502 ± 415 μg/liter) were compared with 17 age-, gender-, and body mass index (BMI)-matched control subjects. The average dialysis duration for the patients was 5.5 years (range = 0.96-18.2 years). Three-dimensional (3D) fast large-angle spin-echo (FLASE) MR images of the distal tibia (voxel size = 137 × 137 × 410 μm3) were processed to yield bone volume fraction (BV/TV). From a skeletonized representation of the trabecular bone network, the topology of each bone voxel was determined providing surface and curve voxel densities (SURF and CURV) and the topological erosion index (EI). Further, high-resolution two-dimensional (2D) spin-echo images were collected at the tibial midshaft for measurement of cortical bone cross-sectional area (CCA), relative CCA expressed as a percentage of total bone area (RCA), and mean cortical thickness (MCT).


The data show both RCA and MCT to be lower in the patients (61.2 vs. 69.1%, P = 0.008, and 4.53 vs. 5.19 mm, P = 0.01). BV/TV and SURF were lower, while EI was increased in the patients, although these differences were not quite significant (P = 0.06-0.09). All of the cortical and trabecular findings are consistent with increased bone fragility.


The data suggest that μ-MRI may have potential to characterize the structural implications of metabolic bone disease, potentially providing a noninvasive tool for the evaluation of therapies for ROD. J. Magn. Reson. Imaging 2004;20:83–89. © 2004 Wiley-Liss, Inc.

BONE DISEASE IN RENAL PATIENTS is a universal finding and is not merely a biochemical abnormality. Common features include bone pain, muscle weakness, extraskeletal calcifications, and fractures. Numerous studies have demonstrated increased fracture rates among end-stage renal disease (ESRD) patients (1–7).

The majority of the studies have been performed with dual-energy x-ray absorptiometry (DXA). However, DXA does not allow distinction of the effects of renal osteodystrophy (ROD) on trabecular and cortical bone. Trabecular and cortical bone behave differently in response to increased parathyroid activity (increase and decrease, respectively). For example, a study of vertebral bone mineral content in women with primary hyperparathyroidism demonstrated increased bone mineral content in the predominantly trabecular vertebral body, but decreased bone mineral content of the cortical spinous processes on lateral spine DXA (8). The conflicting data on DXA-derived measures of bone mineral density (BMD) in ROD are consistent with these limitations. Predictably, DXA results have been variable with mean BMD values that are higher than (9, 10), the same as (11, 12), or lower than those in control subjects (7, 13–18).

Fracture risk does not correlate well with DXA measures of trabecular BMD in patients with renal disease (7, 19). Hemodialysis patients with vertebral fractures and fragility fractures have comparable lumbar spine BMD, compared to those without fractures (7, 19). These observations are not unanticipated, given that trabecular sclerosis in high-turnover disease results in increased DXA BMD despite disruption of structural integrity (1). Cortical bone loss is an early complication of renal disease (20) and is likely to result in significant decrements in bone strength. Consistent with this theory is a report that DXA BMD measures at the cortical radius were significantly lower in spine fracture patients than in those without spine fractures, while lumbar spine measures were not different (19).

Recent advances in high-resolution imaging technologies–notably quantitative computed tomography (QCT) (21) and micro-magnetic resonance imaging (μ-MRI) (for reviews of the subject, see, for example, 22-24)–now enable direct visualization and quantification of both trabecular and cortical bone. μ-MRI has the advantage of being free of radiation and has been shown to provide high spatial resolution. In fact, it is now possible to obtain a three-dimensional (3D) representation of trabecular bone architecture in a volume of interest (VOI) selected from a high-resolution set of contiguous slices, which thus can be analyzed analogous to a physical bone biopsy (23). The method is applicable to peripheral appendicular sites such as the distal radius or tibia, or the calcaneus, where the achievable signal-to-noise ratio (SNR) allows imaging at a voxel size sufficient to resolve the trabeculae. In spite of its demonstrated potential for the study of bone architecture, there is only one report in which μ-MRI was applied to renal disease and, in this case, only to trabecular bone (25). In this pilot study we test the hypothesis that μ-MRI can capture some of the architectural implications of the disease on both trabecular and cortical bone by comparing the results to age- and body mass index (BMI)-matched controls.


Study Subjects

Seventeen patients (nine males) with ESRD treated with maintenance hemodialysis were enrolled. The study was limited to younger adults, age less than 50 years, in order to minimize the confounding effects of aging-related bone loss. Specifically, all female study subjects were premenopausal. Patients with other disorders known to affect bone metabolism were excluded, including diabetes mellitus, systemic lupus erythematosis, and glucocorticoid therapy. Most ESRD patients had persistent secondary hyperparathyroidism (mean parathyroid hormone (PTH) = 502 ± 415 μg/liter). Two patients had low PTH levels (33 and 24 μg/liter), one following parathyroidectomy one year prior. Patients and control subjects were matched for age (controls = 40.2 ± 6.7 years, patients = 40.3 ± 6.4 years, P = 0.9), gender, BMI (controls = 29.5 ± 5.8, patients = 28.5 ± 7.8, P = 0.7), and race (all but two subjects were African American). The average dialysis period for the patients was 5.47 years and ranged from 0.96-18.2 years. For all patient studies, informed consent was sought under an institutionally approved research protocol.


MRI was performed at the right tibia using a Signa™ 1.5-Tesla MRI scanner (GE Medical Systems, Milwaukee, WI) operating in 5.7 system configuration. The protocol comprised scans designed to evaluate both trabecular and cortical structure since both are expected to be affected in ROD. Among the sites at which the VBB for the quantification of trabecular network architecture has been practiced in the authors' laboratory, the distal tibia has the advantage that trabecular structure is relatively homogeneous across the bone's cross section. Further, the tibia ensures superior patient comfort and the foot can more effectively be immobilized than the wrist (another common site for structure evaluation). Imaging was performed with a custom-designed receive-only radiofrequency phased-array surface coil. Subjects were placed supine, feet first, into the scanner. The entire foot was immobilized using a vacuum bag (VacFix™, Soule Medical Systems Inc., Tampa, FL) and straps placed around the lower foot and coil. The coil was placed on the anterior right tibia, using the alignment light of the scanner to place the distal edge of the coil 1 cm distal to the midpoint of the medial malleolus.

A series of axial localizer images was then acquired (field of view (FOV) = 24 × 24 cm2, repetition time/echo time (TR/TE) = 300/14 msec, matrix size = 256 × 128 pixels, number of excitations (NEX) = 0.75), from which the axial slice proximal to the distal tibial end plate was determined. On the basis of the selected axial localizer slice, sagittal high-resolution localizer images were prescribed across the entire width of the tibia. The central sagittal image showing the most proximal location of the cortical end plate was then chosen to prescribe the high-resolution trabecular bone scan, which consisted of a fast large-angle spin-echo (FLASE) 3D image series acquired with the following parameters: FOV = 7 × 5.25 cm2, TR/TE = 80/9.7 msec, matrix = 512 × 384 square pixels, slice thickness = 410 μm (voxel size = 137 × 137× 410 μm3), 28 slices, NEX = 1, flip angle = 140°, scan time = 16.3 minutes. The distal boundary of the scan volume was located 8 mm from the tibia's distal cortical end plate. A typical sagittal high-resolution localizer image of the distal tibia along with one of the cross-sectional high-resolution images is shown in Fig. 1.

Figure 1.

a: Coronal localizer image of the distal tibia with the scan location for acquisition of the high-resolution images placed proximal to the growth plate. b: One of 28 contiguous transaxial FLASE images from which the structural data are derived. Pixel size was 137 × 137 μm2 at a slice thickness of 410 μm.

For the measurement of cortical bone parameters, the standard General Electric transmit-receive extremity coil was used. Axial two-dimensional (2D) fast spin-echo images were prescribed proximal to the distal cortical end plate (31 slices, slice thickness/interslice spacing = 2/5 mm, matrix = 2562, FOV = 12 × 12 cm2, echo train length (ETL) = 8, TR/TE = 5000/16 msec, NEX = 1). This protocol ensured coverage of at least the distal half of the tibia.

Image Processing and Analysis

Trabecular Analysis

The high-resolution FLASE images from the distal tibia were processed by a method similar to that described in Wehrli et al (26) using a custom-designed processing package written in IDL™ (Interactive Data Language, Boulder, CO). The processing steps entailed: 1) correction for subject motion (27), filtering, and Fourier transformation; 2) bone volume fraction mapping (28) to generate noiseless parametric images where each voxel represents the fractional trabecular bone volume (BV/TV); 3) subvoxel processing (29) to improve resolution resulting in a voxel size of 69 × 69 × 103 μm3; 4) binarization and skeletonization (30) yielding a representation of the trabecular network consisting of surfaces and curves (the lower-dimension counterparts of trabecular plates and struts); and 5) topological analysis (31) providing as output the densities of the different topological voxel types (surface and curve types as well as their mutual junctions (26, 30)). Topological parameters examined were the total surface voxel density (SURF, a quantitative measure of the network's platelikeness), the curve density (CURV, the density of all voxels pertaining to a curve, thus a measure of the strutlikeness of the bone), and the topological erosion index (EI, measure of the connectivity of the network). For the exact definition of the topological parameters, see Wehrli et al (26) and Gomberg et al (30). Finally, trabecular thickness (Tb.Th) was measured by the fuzzy distance transform (FDT) method (32, 33). This method allows accurate quantitation of measurement of Tb.Th in the limited spatial resolution regime of in vivo MRI (23).

Cortical Analysis

The cross-sectional high-resolution images of the lower tibial shaft, acquired as described previously, were processed so as to isolate the endosteal and periosteal boundaries. Full details of the method will be reported elsewhere. In brief, boundary detection was based on tracing lines that transect the boundaries of the bone in an approximately perpendicular manner. The intensity profiles along these lines exhibit steep gradients near the boundaries due to the large differences in signal intensity between bone, which has nearly background intensity, and the endosteal and periosteal soft tissues. The MR image intensities along each profile were sampled from the normalized image, and the profiles arranged consecutively to form a 2D array. After the profiles were median filtered to remove noise, the data were binarized at an empirically determined threshold of 30% of the maximum intensity. Finally, the cortical boundary points determined along each profile were mapped back to the original MR image, and the endosteal and periosteal boundaries were thus located. The images from all 28 slices in each study were processed in this manner. The various processing steps are illustrated in Fig. 2.

Figure 2.

Boundary detection for measurement of cortical parameters in MR images of the tibial shaft: raw image (a), radial lines yielding intensity profiles (b), cross-sectional image with boundary highlighted (c), and isolated cortex and marrow (d).

The cortical cross-sectional area (CCA, area between endosteal and periosteal boundaries) and cortical bone area (CBA, area encompassed by the periosteal boundary) were computed as the sum of pixels multiplied by the pixel area. In addition, a relative cortical area (RCA) was computed as CCA/CBA. Mean cortical thickness (MCT) was calculated by modeling the endosteal and periosteal boundaries in each image as concentric circles whose radii were derived from the respective areas. Since bone structural parameters are location dependent, the location along the shaft at which the bone area reached a minimum was used as the location along the bone from which structural parameters were derived.


Cortical and trabecular structural parameters were compared between patients and controls on the basis of two-sided t-tests. Further, possible associations were examined between structural parameters and average PTH levels and dialysis periods using linear regression. Linear regression was also used to evaluate the relationships among structural parameters, age, and weight.


Figure 3 shows a set of images from a typical patient, along with a plot of the cortical parameters as a function of location. Although the appearance of trabecular architecture was found to vary significantly among subjects, patients generally had a less connected network and sparser trabeculation. The group comparisons (Table 1) indicate lower TB/TV, lower topological surface density (SURF), and increased erosion index (EI), although these parameters did not quite reach statistical significance (P = 0.06-0.09). The differences of the cortical parameters were stronger in that the patients had reduced MCT and RCA (P = 0.01 and 0.008, respectively). There were no correlations between average PTH levels or dialysis days with any of the structural parameters, likely due to the marked variability in PTH levels within individual dialysis patients.

Figure 3.

Set of images obtained from a ROD patient (African American, male, age = 42 years) with an average PTH level of 570 μg/L who had been on dialysis for two years: cross-sectional FLASE image from the distal tibial metaphysis (a); 3D rendition of virtual bone biopsy core taken at the site of highlighted circle after processing, showing a relatively disconnected network (b); image at midcalf level (c); and plot of cortical parameters vs. location from distal to proximal direction (d).

Table 1. Means and SDs for Cortical and Trabecular Structural Parameters in Patients and Controls*
  • *

    For abbreviations, see text.


Within the control subjects, body weight was associated with significantly greater CBA (r2 = 0.51, P = 0.001) and CCA (r2 = 0.46, P = 0.003); this association was not seen in the dialysis patients. Also, male control subjects had significantly greater CCA, cortical area, and MCT compared with females (all P < 0.01); again, no gender differences were observed within the dialysis patients.

The substantial variability in cortical and trabecular architecture is evident in the juxtaposition of the images from a high-PTH and a low-PTH patient and their comparison with a control subject (Fig. 4). The high-PTH subject (average PTH = 614 μg/L) shows evidence of marrow fibrosis and trabecular sclerosis (Fig. 4a) and a relatively disconnected network (3D display of virtual core, Fig. 4b), as well as thinned cortex (Fig. 4c). By contrast, the low-PTH patient (average PTH = 33 μg/L) exhibits a widely spaced and sparse network of trabeculae (Fig. 4d and e) and a thin cortex with evidence of endosteal porosity (Fig. 4f), as indicated by intracortical signal. The two cases starkly contrast with the data from a control subject (Fig. 4g-i).

Figure 4.

Comparison of two patients (a-f) with a control subject (g-i); all three African American men. a-c: Patient, age = 40 years, average PTH level = 614 μg/L, total dialysis period = 4.6 years days. d-f: Patient, age = 33 years, average PTH level = 33 μg/L, total dialysis period = 17.4 years. g-i: A 36-year-old control subject. Left column: Cross-sectional high-resolution image (one of 28 contiguous 410-μm slices). Center: 3D surface display of virtual core at the location of the highlighted circle. Right: Image at the location of minimal cortical area in the calf. Note the relatively disconnected trabecular network of the two patients and the very sparse trabeculation in the low-PTH patient. Also noticeable are suspected fibrous marrow inclusions in the high-PTH subject (a, arrow), as well as a thinned cortical shell of the tibia in both patients (c and f).


The key findings of this pilot study relating to the cortical bone manifestations confirm observations reported by Parfitt (34) from iliac crest bone biopsies in secondary hyperparathyroidism, namely, decreased cortical thickness and cortical CCA. Schober et al (35) found the cortical volume measured in the ilium to be reduced in patients with osteitis fibrosa relative to age-matched controls, but peripheral bone densitometry could not distinguish the two groups. Most evidence of cortical thinning stems from iliac crest bone biopsies, although recently data from peripheral QCT (pQCT) have become available in the form of measurements in the forearm (36), suggesting reduced cortical density. These data corroborate earlier measurements in the radius by Di Leo (37), who found both cortical area and thickness measured by pQCT in the distal forearm to be lower in patients with severe secondary hyperparathyroidism than in age-matched controls.

It is interesting in our data that without normalization (CCA vs. RCA) the discrimination between controls and patients was not significant. It is, of course, well known that CCA scales with anthropometric parameters. Indeed, we found weak correlations between CCA and both height and weight (r2 = 0.13 and 0.19, respectively), but there were no correlations with RCA. Further, our data provide evidence of increased porosity, which manifests in the appearance of MR signal in cortical bone, which ordinarily is signal-free (see Fig. 4f). Evidence of increased cortical porosity was previously provided from bone biopsies in patients with untreated primary hyperparathyroidism (38). These authors also reported significant (albeit weak) negative correlations between serum PTH levels and cortical area and thickness, whereas no such associations were observed in the current study. Finally, the notion that cortical thinning occurs through periosteal apposition and enhanced endocortical erosion is suggested by our data, which reveal an increase in total bone area (451 vs. 432 mm2), although this association was not significant. Nevertheless, such an effect is dramatically evident in the two patients whose images are shown in Fig. 4.

Our trabecular bone data suggest structural deterioration in ROD patients in terms of network topology-increased EI and decreased topological surface density, previously established as hallmarks of postmenopausal osteoporosis (26). However, these associations did not quite reach statistical significance, most likely due to insufficient power of this relatively small pilot study, and the relative heterogeneity of the patient population, varying substantially in terms of dialysis duration and average PTH levels. Amling et al (39) reported data from vertebral trabecular bone obtained at autopsy from subjects who had been on maintenance dialysis. Their data suggested that overall BV/TV in patients was similar to that in subjects with intact skeletons, but that trabecular connectivity was often impaired in dialysis patients. Such a phenomenon is also implied by the present data and has been described previously by Hruska and Teitelbaum (40).

Since fibrotic tissue has lower proton density and shortened T2, its signal intensity is considerably below that of fatty marrow and can thus easily be discerned from normal marrow. Nonetheless, while clearly present in both the distal tibia and the midshaft in some subjects (see Fig. 4a and c), not all high-PTH patients show evidence of marrow fibrosis. A histologic hallmark of osteitis fibrosa, found by histomorphometry, is trabecular sclerosis, i.e., a thickening of the trabeculae (40). While some of our data are indicative of such a process, overall trabecular thickness in the patients was not different from that in control subjects.

Reports on the application of 3D imaging in renal disease are scant. Link et al (25) recently reported the results of high-resolution MRI in the calcaneus, QCT in the lumbar spine, and DXA in the proximal femur in patients with renal disease. Receiver operating characteristic (ROC) analysis indicated μ-MRI-derived apparent BV/TF and trabecular spacing (Tb.Sp) to be the strongest differentiators of fracture from nonfracture among renal transplant recipients. Similar values were seen for QCT estimates of BMD in the spine; however, lower values were observed for the DXA results in the proximal femur. A study of trabecular architecture using high-resolution CT in the lumbar spine showed differences in trabecular area, number of trabecular plates, and trabecular diameter in renal transplant recipients, compared with age-matched controls, despite similar BMD (41). Importantly, alterations in bone architecture were associated with progressive vertebral height loss.

Our data, along with the recent high-resolution MR and CT studies discussed above, underscore the utility of assessment of trabecular and cortical architecture in renal disease. Since renal disease is associated with marked cortical bone atrophy, with consequent reductions in strength (34), the ideal modality for the assessment of fracture risk in renal disease will incorporate measures of trabecular and cortical architecture, as described in this article.

In conclusion, this pilot study suggests that quantitative high-resolution MRI can provide detailed insight into the architectural implications of metabolic bone disease affecting both cortical and trabecular bone. The method will require further validation in larger patient studies to determine its ultimate clinical potential for supplanting invasive biopsy in the evaluation of patients with renal disease.


The authors acknowledge Louise Loh for her assistance in patient recruitment.