Osteodensitometry of human heel bones by MR spin-echo imaging: Comparison with MR gradient-echo imaging and quantitative computed tomography



The aim of the study was to investigate whether quantitative magnetic resonance (MR) fast spin-echo (FSE) imaging with moderate spatial resolution enables osteodensitometry in peripheral yellow bone marrow. Signal intensities in T1-weighted FSE images from yellow bone marrow indicate the amount of adipose tissue per volume. The signal intensity in marrow regions with spongy bone was assessed and compared to signal intensity of pure fatty marrow (100%). Heel bones of 30 patients with suspected osteoporosis were analyzed and the FSE images were compared with results from parallel MR gradient-echo (GE) imaging and quantitative computed tomography (QCT) examinations. High correlation was found between FSE imaging and QCT [r = 0.91 in the dorsal region of interest (ROI); r = 0.86 in ventral ROI]. Linear correlation coefficients between GE imaging and QCT were slightly lower in the dorsal part (r = –0.86) and considerably lower in the ventral part (r = –0.68). Correlation between the two MR techniques amounted to r = –0.72/–0.61 (dorsal/ventral). The high correlation between FSE imaging and bone mineral density (BMD) allows possible clinical applications of FSE imaging for diagnosis of osteoporosis. Further improvements of the accuracy using reference phantoms might be possible. J. Magn. Reson. Imaging 2001;14:147–155. © 2001 Wiley-Liss, Inc.

OSTEOPOROSIS IS A SYSTEMIC skeletal disease characterized by a loss of both trabecular and cortical bone substance. Mechanical properties of the bone might change dramatically and lead to (atraumatic) fractures, mainly affecting the spine (vertebral bodies), hip, wrist, and ribs (1). Quantitative computed tomography (QCT) and dual energy X-ray absorptiometry (DEXA) are accepted as gold-standard modalities for osteodensitometry providing a reliable measure of bone mineral density (BMD) of trabecular bone (2–4) or areal bone mineral density (BMAD) (5, 6). Unfortunately, both modalities lead to radiation exposure to the patients. Great effort has been made over the past years to develop imaging modalities for non-invasive quantitative assessment of bone density and/or the spatial distribution of trabecular structures using ultrasound (7, 8) and magnetic resonance (MR) (9–17) techniques. MR imaging has been reported to offer two different approaches for the assessment of density and/or distribution of trabecular structures: >

  • 1Techniques sensitive to the microscopic magnetic field distribution [gradient-echo (GE) imaging (12, 13), asymmetric spin-echo imaging (14), or phase imaging (15)] allow an estimation of the structure and density of the trabecular network by determination of the microscopic magnetic field inhomogeneity, which is influenced by the different susceptibilities of bone marrow and trabecular structures.
  • 2High resolution (HR) MR techniques have been applied for direct visualization of the trabecular network (16, 17), providing an excellent insight into the architecture of spongy bone. The relatively long measurement time of about 15 to 20 minutes to obtain a sufficient signal-to-noise ratio (SNR) is disadvantageous. In addition, inevitable movement artifacts in patient examinations have led to limited clinical applicability of highly resolved imaging techniques.

Over the past years, several MR techniques were applied in vitro and in prospective pre-clinical studies in order to find reliable tools for the evaluation of pathologic osteoporotic changes without radiation exposure. Sufficient correlation between MR examinations and the gold standard have been reported for yellow bone marrow (12, 13, 18, 19). In contrast, the conditions in red marrow have been shown to be clearly more complex, since the composition of marrow components (including paramagnetic hemosiderin) is variable and of influence on MR osteodensitometry (20–23).

In adults, peripheral bone marrow (e.g., in the heel bones) consists nearly exclusively of fatty tissue after the bone marrow conversion during the first two decades in human life. Former bone marrow investigations using proton MR spectroscopy and chemical shift imaging confirmed that no significant signals other than fat became visible from the foot skeleton of adults (11, 23) if inflammatory processes and generalized hematological disorders were excluded.

The aim of this study was to investigate whether simple MR fast spin-echo imaging (FSE) with moderate spatial resolution offers an alternative to QCT for the assessment of BMD in peripheral bone. The method is based on the well-known fact that compact bone and trabecular structures yield no signal in MR imaging because of the solid material and the resulting extremely short transverse relaxation times. Thus, the detected signal in the picture elements (pixels) stems exclusively from bone marrow, and a reduced signal intensity compared to pure fatty tissue indicates the presence of bony structures. The results of FSE imaging were compared to those of GE imaging sensitive to the susceptibility related T2* effects from the same regions. Furthermore, BMD of the selected marrow regions was measured by QCT, allowing a direct comparison of the MR imaging modalities with a gold standard. Heel bones were chosen as the location for the osteodensitometric measurements considering the fact that they consist exclusively of fatty bone marrow and are good predictors of the fracture risk in patients suffering from osteoporosis (8).


Examinations of the heel bones were performed in thirty consecutive adult patients (18 male, 12 female, ages 31–65 years) that were suspected to suffer from osteoporosis because of inflammatory bowel disease and treatment with corticosteroids. Patients with generalized skeletal diseases, hematological diseases, inflammatory bone marrow processes, or old fractures in heel bones were excluded from the study. All included patients gave written informed consent. The study design was approved by our local ethics committee.

MR and QCT investigations were performed on the same day in our hospital. In order to minimize movement artifacts and to enable identical positioning in MR and CT scanners, the feet of the patients (in supine position) were fixed in 15° plantar flexion during all examinations on a special MR- and CT-compatible foot cast made of Perspex®.


MR measurements were performed on a 1.5-T standard whole body unit (Magnetom Vision, Siemens, Erlangen, Germany) using a circularly polarized headcoil (transmitter and receiver coil). The protocol included FSE and GE imaging. Transverse slices were recorded from the distal half of the heel bones after positioning on a sagittal slice. The complete measurement time for the MR-protocol amounted to about 20 minutes.

FSE Imaging

A common T1-weighted FSE sequence was applied for the examination of the heel bones. Measurement parameters were TE = 12 msec, TR = 650 msec, 3 scans, matrix: 384 × 512, field of view (FOV): 230 mm × 307 mm. Three echoes were measured after each excitation in the FSE train. The pixel size was 0.6 mm2 with a slice thickness of 2 mm. Thirteen parallel slices were recorded in transverse orientation. Acquisition time amounted to 4 minutes and 16 seconds. Signal intensity of pure fatty tissue served as reference.

Correction of Inhomogeneous Sensitivity of the Receiver Coil

Despite the relatively smooth sensitivity characteristics of the used volume coil (circularly polarized head coil of the manufacturer), signal intensities from homogeneous phantoms showed slight inhomogeneities at the rim of the coil.

Without correction, these inhomogeneities of sensitivity would cause a systematic error in the interpretation of the signal intensity from spongy areas of heel bones. For a reliable quantification of the volume of spongy bone (trabecular share, TS) as a percentage, a correction of the signal intensities was applied (see Fig. 1). For this purpose, a phantom measurement was carried out using a NiSO4-doped aqueous solution in a Perspex sphere with a diameter of 25 cm. The phantom was centered within the head-coil and images were recorded with identical measurement parameters as applied in vivo (Fig. 1b).

Figure 1.

Signal correction of FSE images. a: Original T1-weighted FSE image of the heel bones of a 51-year-old male patient. b: Phantom image used for signal correction, showing the signal inhomogeneities on the edges of the recorded FOV. c: Corrected FSE image used for signal postprocessing. d: Difference image between original (a) and corrected (c) FSE images. Bright regions indicate an increase of signal intensity due to the correction.

Signal correction of the evaluated regions of interest (ROIs) was performed by dividing the mean signal intensity (SI) from the selected ROI (SImeas) by the mean SI of the corresponding ROI in the phantom image (SIphantom) according to the equation:

equation image

A noise correction was not necessary since the obtained SNR in phantom and patient measurements was clearly higher than 10.

For evaluation of the signal intensity of pure fat in the FSE images, the maximum signal intensity in bone marrow was measured in a subregion of a few pixels in the ventral part of the calcaneus, which normally shows large caverns without trabeculae. This intensity value was corrected with the corresponding phantom data and then set to 100% fat, serving as an internal reference for the assessment of the TS.

GE Imaging

GE images were recorded from the same slice positions as in FSE imaging with an in-plane resolution of 0.6 mm2. A three-dimensional GE multi-echo imaging technique was applied, recording two images with echo times of 9.3 msec (TE1) and 27.9 msec (TE2). Echo times were chosen considering the in-phase condition of water and methylene signals at 1.5 T. Fifteen partitions with a thickness of 0.6 mm were recorded with a repetition time (TR) of 100 msec and a flip angle of 45°. Four successive partitions were added resulting in a representation of a 2.4 mm slice for the comparison with corresponding FSE images with 2.0 mm nominal slice thickness. The measuring time for the three-dimensional GE sequence with 1 scan amounted to 10 minutes and 16 seconds. Signal decrease in GE images for prolonged echo times was determined as signal ratio Q = S(TE2)/S(TE1) of the noise-corrected signal intensities (SIcorr). The signal ratio Q was chosen instead of a two point exponential fit, since the signal decrease in bone marrow is known to show a deviation from a monoexponential behavior (24). Noise correction was performed by

equation image

where SIROI is the mean signal intensity of the chosen ROI, and N is the mean signal of noise determined from a ROI in an object-free region of the image (25). The square values in the formula are necessary due to the complex data type obtained after Fourier transformation of the raw data.


QCT examinations were performed on a Somatom Plus IV (Fa. Siemens, Erlangen, Germany). Images were recorded in planparallel transverse slices without gantry tilt. Slice thickness was chosen as 1-mm with an acceleration voltage of 120 kV. For calculation of the BMD, a standard phantom for osteodensitometry with different shares of calcium-hydroxyl-apatite (CaHA) was positioned below the feet of the patients and measured simultaneously. Mean Hounsfield units (HU) of water (0 mg CaHA/mL) and a solution containing 250 mg CaHA/mL were determined. Signal intensity of fat (HUfat = –90 ± 10 HU at calibrated units) was calculated by linear adaption and set to BMD = 0 mg CaHA/mL for determination of BMD values in the heel bones. For evaluation of the trabecular density, two QCT-slices were averaged to obtain an effective slice of 2 mm as in corresponding FSE images. The rationale to select a smaller slice thickness in QCT than in FSE imaging was to have a higher flexibility in the generation of a CT image with an optimal correspondence to the MR images.

Evaluation of Trabecular Bone Content

For assessment of the trabecular bone content by FSE, GE, and QCT images, transverse sections of the most extensive part of the heel bones were selected and displayed side by side on a separate workstation. From each heel bone, two circular ROIs of approximately 1.8 cm2 were selected, one in the dorsal part and one in the ventral part of the calcaneus. The dorsal region is usally characterized by a high amount of a close-meshed trabecular network, whereas in the ventral region, only a few trabecular struts are present with large spaces of pure fatty bone marrow between them. Inclusion of cortical bone was carefully avoided in all selected ROIs. Mean signal intensities were determined from the ROIs in all selected images and, additionally, from a region in an object-free part of the GE image for the noise correction mentioned above.

Statistical Analysis

Statistical analysis was performed using SigmaStat® software tools (Jandel Scientific). Linear regression analyses were performed to evaluate the correlations between the TS values from FSE imaging, the derived signal ratios Q from the GE images, and the BMD values from QCT examinations. Differences between distal and proximal ROI were analyzed using a paired Student's t-test. A P value of less than 0.05 was considered statistically significant.


All patients were able to undergo both examinations by MR imaging and QCT. Quality of images was acceptable in all cases and applied foot casts allowed a very good correspondence between MR and QCT images. Results from the single imaging methods and correlations between the three imaging modalities are described below.

MR FSE Imaging

Examples of intensity-corrected FSE images of the heels of a 45-year-old male patient with normal BMD (a) and a 62-year-old female patient with severe osteoporosis (b) are exhibited in Figure 2. The signal intensity in the dorsal part of the heel bone was clearly higher in the osteoporosis patient compared to the normal case. This finding is quantitatively expressed in the TS of only 13.6% in the right heel bone (RH) and 13.8% in the left heel bone (LH) in the patient with osteoporosis, compared to 32.4%/33.2% TS (RH/LH) of the male patient with normal BMD.

Figure 2.

FSE images of the left heel bone of a 45-year-old male patient with normal trabecular density (a) and a 62-year-old female patient with reduced trabecular density (b). Typical positions of the ROIs in the dorsal (d) and ventral (v) part of the heel bone are depicted in (b).

In the dorsal ROI, TS values of all patients were in a wide range between 13.6% and 41.4% (28.0 ± 5.7%), corresponding with the variable age and status of the patients. TS values in the ventral ROI were significantly lower (14.6 ± 5.6%, P < 0.0001), in accordance with the usually lower trabecular density in this site.

MR GE Imaging

Figure 3 presents two sets of GE images of the patient with normal trabecular density (a–c) and the patient with severe osteoporosis (d–f). Echo times are TE1 = 9.3 msec (a,d) and TE2 = 27.9 msec (b,e). Figures 3c and 3f depict parameter images, calculated as signal ratio of images with TE2 and TE1.

Figure 3.

GE images of the left heel bone of the 45-year-old male patient with normal trabecular density (ac) and the 62-year-old female patient with reduced trabecular density (df). Echo times were TE = 9.3 msec (a,d) and 27.9 msec (b,e). The parameter images (c and f) depict the calculated signal ratios Q of the entire images.

In the dorsal ROIs, the relative decrease of signal intensity for longer echo time was clearly more pronounced in the normal case (Q = 0.41) compared to the osteoporotic patient, who showed a bright signal in this region in the image with echo time TE2, resulting in a signal ratio of Q = 0.68. This also becomes obvious from the parameter images, where bright regions indicate a slow signal decrease. In the ventral ROI, the signal decrease was less pronounced for all subjects with a significantly higher mean signal ratio Q (P < 0.0001).

The signal ratio Q for the entire cohort of patients amounted to 0.48 ± 0.07 in the dorsal part and 0.66 ± 0.09 in the ventral part of the heels.


In Figure 4, the QCT images of the patient with normal BMD (a) and the osteoporotic patient (b) are shown. A very high congruence of the selected slices in the MR and CT examination was obvious. The osteoporotic patient showed a clearly darker signal in the complete spongy area compared to the patient with normal trabecular density. BMD was 196.5 mg CaHA/mL for the patient with normal BMD and 48.3 mg CaHA/mL for the patient with reduced BMD.

Figure 4.

QCT images of the left heel bone of the 45-year-old male patient with normal trabecular density (a) and the 62-year-old female patient with reduced trabecular density (b).

BMD values calculated from QCT images by linear transformation of HU showed significant differences between the dorsal and ventral ROI in all patients (dorsal: BMD = 162 ± 38 mg CaHA/mL, 48 < BMD < 254; ventral: 73 ± 38 mg CaHA/mL, 0 < BMD < 149; P < 0.0001).

Correlation Between the Imaging Techniques of MR and QCT

A comparison between FSE, GE, and QCT was performed. Table 1 lists the calculated values from dorsal and ventral ROIs of the three techniques. In FSE imaging, a significantly higher TS resulted in the dorsal ROI. Correspondingly, the signal ratio Q determined by GE imaging was significantly lower and BMD values were significantly higher in the dorsal part of the calcaneus (P < 0.0001 each). Furthermore, differences between the right and left heel bones were examined. Results are given in Table 1. All methods indicated a slightly higher trabecular density (expressed as TS, Q, or BMD) in the right heel compared to the left heel. However, differences were not significant in all three techniques (P > 0.05).

Table 1. Quantitative Data From Fast Spin Echo Imaging (FSE; TS), Gradient Echo Imaging (GE; Q), and Quantitative Computed Tomography (QCT; BMD)
Selected ROIsFSE imaging trabecular share (TS)GE imaging signal ratio (Q)QCT bone mineral density (BMD)
Mean ± SDRangeMean ± SDRangeMean ± SDRange
Dorsal (all ROIs)28.0 ± 5.713.6–41.40.48 ± 0.070.35–0.68161.7 ± 38.448–254
Ventral (all ROIs)14.6 ± 5.64.7–32.90.66 ± 0.090.43–0.8772.6 ± 38.30–149
Dorsal (left ROIs)27.3 ± 5.313.8–39.40.49 ± 0.070.38–0.68156.0 ± 38.460–254
Dorsal (right ROIs)28.6 ± 6.113.6–41.40.48 ± 0.070.35–0.67167.5 ± 38.248–244

Correlations between FSE, GE, and QCT were calculated for all evaluated ROIs (rall), and for the separated ROIs in the dorsal (rdor) and ventral part (rven) of the heels using linear regression analysis.

The highest correlations were seen between TS and BMD in both regions examined. Correlation coefficients amounted to rall = 0.94, rdor = 0.91, and rven = 0.86. Results are shown in Figure 5. Correlation between Q and BMD was slightly lower in dorsal region with rdor = –0.86 and clearly lower in ventral region, with rven = –0.68. Correlation for all regions evaluated amounted to rall = –0.88 (Fig. 6). Comparison between values determined by FSE and GE imaging resulted in rall = –0.85, rdor = –0.72, and rven = –0.61, as shown in Figure 7.

Figure 5.

Correlation between BMD measured by QCT and TS calculated from FSE imaging. a: All ROIs. b: Dorsal ROIs. c: Ventral ROIs.

Figure 6.

Correlation between BMD (QCT) and signal ratios Q from GE examinations. a: All ROIs. b: Dorsal ROIs. c: Ventral ROIs.

Figure 7.

Correlation between TS (FSE) and signal ratios Q (GE). a: All ROIs. b: Dorsal ROI. c: Ventral ROI.


In contrast to techniques sensitive to the microscopic magnetic field distribution, FSE sequences provide completely rephased signals contributing to the recorded echo. Thus, the proton density (i.e., the relative volume share of fatty tissue) exclusively determines signal intensity in FSE images of fatty bone marrow if the relaxation behavior is constant for all fatty marrow sites (26–28). Consequently, information about the content of trabecular structures in a defined ROI in yellow marrow can be obtained by analysis of the relative signal intensity reduction in spin-echo images. The presented data demonstrate that MR FSE at moderate spatial resolution offers a non-invasive tool for the assessment of the trabecular fraction in peripheral bone marrow regions without radiation exposure to the patients. Our examinations of peripheral bone marrow in adults led to an average error of approximately 15 mg CaHA/mL and a maximum error of 45 mg CaHA/mL in a comparison with the BMD obtained by QCT (serving as gold standard).

A linear correlation between TS and BMD was expected, since both methods measure the amount of trabeculae, without an influence of the spatial arrangement of trabecular structures. Thus, a clinical application of the FSE MR method instead of QCT seems to be feasible. Only one image has to be recorded for each slice and postprocessing of the FSE images is relatively simple compared to GE images, where at least two images have to be recorded and postprocessed. The reference scan from a homogeneous phantom for corrections of inhomogeneous coil characteristics can be applied for all patients if slice parameters remain unchanged.

The correlation with the gold standard was clearly better for FSE imaging than for susceptibility-based GE imaging. This finding is probably due to the fact that FSE only depends on the amount of trabecular material, whereas GE imaging additionally depends on the spatial arrangement and orientation of the bony structures. Different magnetic susceptibilities of bone marrow (fat) and trabecular structures cause inhomogeneities of the magnetic field at the transitions between these compartments, leading to signal dephasing and increasing signal loss at increasing echo times (29). Especially in the ventral ROI, the correlation between FSE and QCT was considerably higher than the correlation between GE and QCT. This might be caused by the structure and orientation of a few thicker trabecular struts streaking the bone marrow in this location and resulting in other susceptibility effects than a close-meshed network with thin trabeculae and with the same BMD. As the type of the correlation between the signal ratios Q and the pure bone density is unknown and dependent on the structure, a linear correlation approach was also applied for the comparison of GE imaging with the other modalities.

FSE and GE imaging (both with moderate spatial resolution) could be used as complementary techniques: FSE provides a measure of the bony mineral content, whereas parallel GE imaging reveals microscopic structural features of the spongiosa. However, it has to be determined which of these features are really important for the mechanical properties of the spongiosa and for the risk of fracture in the patients.

There are some limitations of FSE imaging concerning the accuracy of bone density measurements. One problem is the exact determination of the signal amplitude for pure marrow tissue without trabeculae. Only a few pixels from the ventral part of the heel bone are expected to contain no trabecular structures. It is difficult to ensure the correctness of this internal reference. Thus, a further step to increased accuracy could be the use of an external reference, but temperature effects and the different relaxation behavior of the reference substance and yellow bone marrow must be considered. A second problem is the fact that even yellow bone marrow does not exclusively consist of bone and pure fat. Low contributions of intra- and extracellular water (including vascular structures) and even further connective tissue without calcification need space in the marrow and lead to reduced signal in FSE images. It must be examined in further studies whether these effects are individually variable and which maximal errors have to be considered. It should be mentioned that influences from pathological composition of the marrow (inflammation, hematological diseases) were excluded in our patients.

The proposed technique probably works reliably in spongy regions with yellow bone marrow with “normal” composition (nearly exclusively fatty content). For this reason, heel bones were chosen for our examinations. As shown by several authors, bone mineral density of heel bones provides information on the bone changes caused by osteoporosis and can serve as an indicator for the estimation of the risk of fractures (e.g., 4,8). However, in young children, even peripheral marrow sites show hemopoetic cells with considerable water content. The same problem is obvious in the axial skeleton of all patients, with markedly variable marrow composition (21, 22). Since the signal intensity depends on the marrow composition, an assessment of the trabecular volume share becomes clearly more difficult in proximal areas with red bone marrow (23). For this reason, clinical applications of the proposed spin-echo method in its present form are only possible on peripheral skeletal sites of adults, where radiation exposure is less problematic than in children. Sequence parameters with very similar signal yield from red and yellow marrow fractions could help in these cases. Furthermore, suitable (internal) reference signals are difficult to assess, since all marrow regions of bones in the axial skeleton contain spongious material.

Another aspect in clinical osteodensitometry is cost-efficacy. In this context, it should be mentioned that there is a marked difference between MR osteodensitometry and QCT and DEXA osteodensitometry, respectively. Taking into account the costs for personnel and maintenance of the scanners, MR-osteodensitometry with a duration of approximately 35 minutes (including FSE imaging and GE imaging) is about two to three times more expensive than a comparable osteodensitometry by QCT in approximately 20 minutes, and clearly more expensive than DEXA. However, the potential of MRI concerning the indirect assessment of the trabecular structures and the missing radiation exposure could warrant the additional financial effort.

There is still a lot of work to do until MR osteodensitometry can be routinely used, but the FSE imaging technique and the proposed data processing leads to reasonable results and could become a basic tool in the protocol for MR osteodensitometry.


The authors gratefully thank the members of Siemens Medizintechnik (Siemens AG, Erlangen, Germany) for technical assistance.