Age, gender, and skeletal variation in bone marrow composition: A preliminary study at 3.0Tesla




To evaluate the efficacy of MR Spectroscopy (MRS) at 3.0 Tesla for the assessment of normal bone marrow composition and assess the variation in terms of age, gender, and skeletal site.

Materials and Methods

A total of 16 normal subjects (aged between eight and 57 years) were investigated on a 3.0 Tesla GE Signa system. To investigate axial and peripheral skeleton differences, non-water-suppressed spectra were acquired from single voxels in the calcaneus and lumbar spine. In addition, spectra were acquired at multiple vertebral bodies to assess variation within the lumbar spine. Data was also correlated with bone mineral density (BMD) measured in six subjects using dual-energy X-ray absorptiometry (DXA).


Fat content was an order of magnitude greater in the heel compared to the spine. An age-related increase was demonstrated in the spine with values greater in men compared to female subjects. Significant trends in vertebral bodies within the same subjects were also shown, with fat content increasing L5 > L1. Population coefficient of variation (CV) was greater for fat fraction (FF) compared to BMD.


Significant normal variations of marrow composition have been demonstrated, which provide important data for the future interpretation of patient investigations. J. Magn. Reson. Imaging 2007;26:787–793. © 2007 Wiley-Liss, Inc.

BONE MARROW consists of both hematopoietic (red) and fatty (yellow) components, the proportions of which are thought to be related to the remodeling capacity of bone. Furthermore, bone strength has been shown to depend not only on bone mineral density (BMD) but also marrow quality (1). Both of these factors have implications in osteoporosis, although it is still unclear why some people are susceptible to osteoporosis and what role the marrow plays. There is a well-established age-related conversion of red to yellow bone marrow, a process that occurs first in peripheral and later in the axial skeleton (2). At any age a sustained increase in demand can lead to the reappearance of hematopoiesis in the extremities. Clearly, a complete characterization of all normal patterns and distributions of marrow is essential before pathologic conditions can be identified.

MRI has become the noninvasive imaging modality of choice in diagnosing bone marrow disorders (3). Red and yellow marrow is readily distinguishable and marrow composition may be qualitatively assessed from signal intensity variations on MRI. Previous work has demonstrated the age-related conversion in cranium (4), femur (5), vertebral bodies (6), epiphyses (7), and pelvis (8). In particular, Ricci et al (6) described three distinct signal intensity patterns in the lumbar spine that changed with age.

In comparison, relatively few MRI studies have examined quantitative measurements of bone marrow, inferring red and yellow composition from their water and fat signal contribution. This is typically done using MR spectroscopy (MRS) whereby the fat and water signal can be acquired from a small voxel. Studies to date have tended to focus on a single vertebral body (VB) in the lumbar spine, usually L2 or L3 (9–11). These studies were also able to demonstrate the age-related increase in fat content.

The idea of gender differences in marrow content is less well established. Although Schellinger et al (10) and Kugel et al (11) demonstrated an increase in fat content in men using MRS, these results remain at odds with the many more imaging studies that have not observed any variation (6, 12, 13).

Schellinger et al (14, 15) subsequently showed a link between bone marrow fat and BMD in osteoporosis. Here, single BMD and MRS measurements were taken as representative of the whole lumbar spine. For fat content to be used as a routine diagnostic parameter, a thorough characterization of all normal variation needs to be understood and should include possible intervertebral differences.

With the advent of clinical high-field scanners, there is an opportunity for combining high spatial resolution imaging with high-sensitivity spectroscopy. The study is divided into three sections: first, in vivo measurements in both axial and peripheral skeletal sites for the same individuals are obtained to quantify the absolute differences. Second, additional subjects were examined in more than one lumbar VB using MRS and two imaging techniques. This data also provides further supportive evidence for the differences in fat fraction (FF) between male and female subjects in the lumbar spine. Results also show a significant VB variation that, to our knowledge, has not been reported previously. Finally, FF data is compared to measurements of BMD in six subjects using the commonly accepted method of dual-energy X-ray absorptiometry (DXA). These results contribute further valuable data to help quantify and understand the normal composition variations of bone marrow.


A total of 16 healthy subjects were examined aged between eight and 57 years (mean 33 years), none of whom had a history of lower back problems. Informed consent was obtained from each subject or parent. In seven subjects, MRS data was acquired in both the calcaneus and a single VB of the lumbar spine (L3) in two separate examinations on the same day. In 10 cases, the lumbar spine was examined at more than one location: two were examined at L3 and L4, four at each of L1, L3, and L5, and a further four cases had spectra acquired in all five VBs (L1 to L5). Voxel location was chosen by the investigators and a final classification was given by a trained radiologist at the Centre. This lead to the subsequent reclassification of the youngest VB measurement from L3 to L4.

All MR examinations were performed on a 3.0 Tesla whole-body GE Signa system (HDx platform) using either a commercially available quadrature knee or CTL cervical, thoracic, lumbar (lower three elements) phased-array coil. For heel examinations, subjects were restrained in the knee coil with padding. For spine examinations a knee pad was used to reduce lordosis. Initial localizing scans were followed by a standard clinical sequence to prescribe the spectroscopic voxels. In the calcaneus, this consisted of a coronal T1-weighted fast spin-echo (FSE) sequence (TE/TR = 12 msec/600 msec) with a 2 mm/0.2 mm slice thickness/gap and 14 × 8 cm field-of-view. For the spine, a sagittal T2-weighted FSE sequence was used (TE/TR = 106 msec/4000 msec) with a 3 mm/1 mm slice thickness/gap and 30-cm field-of-view.

Spectroscopy data was acquired using the point-resolved spectroscopy sequence (PRESS) with TE/TR = 35 msec/2000 msec. Non-water-suppressed spectra were obtained by setting water suppression voltages and flip angles to zero. A single voxel was prescribed from a region of approximately 1.5–2.0 cm3 in size within trabecular bone at each particular site or VB. For the lumbar spine, outer volume saturation bands were used to eliminate unwanted signal contamination from outside the voxel. These typically covered the cerebrospinal fluid (CSF) and adjacent vertebral discs. After local shimming and gradient adjustment, a total of 64 or 128 signals were collected with a spectral bandwidth of 5000 Hz and 4096 data points. Examination times varied from a minimum of 20 to a maximum of 45 minutes for a complete investigation of all five VBs.

Data was analyzed offline using the SAGE package (GE Medical Systems) and processing included: zero-filling, spectral apodization (using a Gaussian function with a 2.5 Hz linewidth), first order automatic phasing, and direct current (DC) baseline correction. Peak amplitudes of the water resonance and the main lipid peak at 1.3 ppm were measured. Percentage FF was calculated from the fat-to-water peak ratios (FWR) according to 100 × FWR/(FWR + 1). The presence of other (lipid) resonances was also recorded (although measurements were not taken) with peak assignments given as I to V following previous work (16).

In the subjects in whom at least three lumbar vertebral bodies were examined, FF was additionally measured using two imaging techniques. Firstly, standard in-phase and out-of-phase imaging (IOP) was acquired following a methodology previously used in the liver (17). This consisted of two fast gradient echo sequences with TE either in (2.1 msec) or out (3.2 msec) of phase and flip angles of 20° and 70° to identify the dominant signal component. This resolves the ambiguity of the signal produced by either water–fat or fat–water that is inherent with magnitude reconstructed images. The second method used a T2-weighted iterative decomposition of water and fat with echo asymmetric and least-squares estimation (IDEAL)-FSE sequence (TE/TR = 60 msec/3000 msec) (18), which produces separate water-only and fat-only images as part of the scan protocol. All images were processed offline using in-house developed software (MATLAB). For the IOP method, this involved calculating FF using both sets of flip angle images with the final value determined from a comparison of the two results; using either FF or 100 – FF as appropriate (17). In the case of IDEAL, FF was calculated from the ratio of the separate water-only and fat-only images. Pixel-by-pixel color-scale maps of FF were produced in both cases and measurements of mean pixel values were then taken from regions-of-interest (ROIs) drawn within each vertebra for comparison with MRS data.

Additionally, in six of these subjects, BMD (g cm–2) was measured at the five lumbar vertebrae using a GE Progidy DXA scanner in standard lumbar spine anterior-posterior scan mode. An automated tissue-typing algorithm determined the ROI measurement for each vertebra.

Two subjects had more than one MRS investigation: In one case (female aged 57 years) spectroscopy measurements in L3 were repeated five months apart as an indication of the repeatability of the technique. In one further subject (male aged 34 years) it was possible to additionally acquire L3 data on a 1.5 Tesla system (GE Signa) for comparative purposes. The same acquisition parameters (with 2500 Hz/2048 points for consistent spectral resolution) and processing method was followed.

All statistical analyses were carried out using SPSS 13.0 (SPSS Inc., Chicago, IL, USA). A univariate analysis of covariance (ANCOVA) was used to determine if sex and age (as a covariate) were significant factors in determining FF. The Friedman nonparametric test (for k related samples) was used to determine if FF was different in the three bones studied (L1, L3 and L5). Analysis was therefore limited to those seven cases where data were available for all three bones; thus using each case to self-correct for age- and sex-related variations. Multiple Wilcoxon paired-rank tests were then carried out in order to investigate the differences between different pairs of bones. Imaging and MRS values were correlated with a Pearson correlation coefficient.


Good quality spectra were acquired in all 16 cases. Figure 1 shows example spectra from the calcaneus and lumbar spine. Figure 1a is a spectrum taken from the calcaneus of a subject aged 13 years showing very little water in relation to fat. Percentage FF was found to be 94.1%. The corresponding spectrum in the spine (L3) of this subject, given in Fig. 1b, shows a marked decrease in fat content with FF = 27.0%. Figure 1c and d show spine spectra taken in subjects aged 28 and 57 years, demonstrating an increase in the relative fat content with age, giving FF values equal to 49.0% and 68.9%, respectively.

Figure 1.

Example spectra illustrating the marked difference in fat content between heel and spine together with age-related changes. a: Shows a single voxel measurement in the calcaneus of a 13-year-old male subject. The corresponding spectrum in the spine (at L3) for this subject is shown in (b). Also shown are example spectra taken in the spine of a 28-year-old male (c) and a 57-year-old female (d), demonstrating the increase in fat content with age. FF values are given in the text.

In addition to the water and main lipid resonance (labeled as peak II in Fig. 1 and corresponding to the methylene chain), at least one other resonance was seen in all the spectra and these are assigned according to previous work (16). In all the spine data a peak at 2.2 ppm was present (peak III in Fig. 1), corresponding to methylene protons attached to the carboxyl chemical group. In spectra where the FF was approximately 57% or greater (based on the 1.3 ppm resonance), i.e., the older spine subjects and in all heel data, a resonance at 5.2 ppm was observed, (peak V in Fig. 1). This corresponds to olefinic protons and the CH group of glycerol. A resonance at 2.5 ppm (peak IV), corresponding to diallylic CH2, was seldom observed and a peak at 0.9 ppm (terminal methyl) was not present in any spectrum.

Figure 2 plots percentage FF in all seven subjects for which heel (squares) and spine (triangles) data were acquired. This clearly shows a large difference between the relative amounts of fat at these two anatomical sites. Fat content increased significantly with age in the spine with FF values varying from 27% to 70% and FWR ranging from 0.37 to 2.27 (P = 0.003). Corresponding values in the heel were much higher, ranging from FF = 83% to 98% and FWR = 16.67–50.00, and these did not correlate with age (P = 0.140).

Figure 2.

A plot of percentage FF vs. age for subjects in which both the heel (squares) and spine (triangles) were examined. There is a significant age-related increase in fat content in the spine but not in the heel, which has a markedly higher fat content even at the earliest age studied.

In Fig. 3, a plot of FF vs. age is given for spine data only and includes the subjects in which more than one VB was examined (16 subjects in total). The data is subdivided into male (blue) and female (pink) subjects with vertebrae denoted by the following markers: L1 (open square), L2 (solid square), L3 (open triangle), L4 (solid triangle), and L5 (star). There is considerable variation, with FF values ranging from 14.3% (L1, female aged 30 years) to 71.1% (L5, female aged 57 years). The graph also shows the overall increase in FF with age, and in addition a tendency of female subjects to exhibit lower values of FF compared to males of similar age. Results of the ANCOVA test revealed statistically significant age (P = 0.008) and gender (P = 0.037) variation in L3 content. There is also a general trend toward increased fat for more inferior VB although this is not seen in every subject. Mean ± SD values of FF in the VBs most studied were as follows: 40.5 ± 16 % (L1), 49.6 ± 13.6 % (L3), and 51.3 ± 13.3 % (L5). Differences both within and between subjects were significant using the Friedman test (P = 0.001) and Wilcoxon test (P = 0.018), respectively.

Figure 3.

A plot of percentage FF for all spine data split into male (blue) and female (pink) subjects. Data points are further classified by the following markers: L1 (open square), L2 (solid square), L3 (open triangle), L4 (solid triangle), and L5 (star). Note: two males aged 34 years: five VBs from one, and a separate higher L3 measurement in a second.

IOP and IDEAL imaging was also performed to obtain an independent estimate of FF. In Fig. 4, a map of FF produced from the IDEAL images is shown in the 57-year-old female subject covering the vertebrae from T12 to L5. The color scale demonstrates increasing fat content from superior to inferior VBs (scale shown is red for lower fat content and yellow for higher fat content). On the whole, FF was overestimated when imaging values were compared to those obtained with MRS. A plot of FF determined by the two imaging methods compared with MRS revealed that IDEAL values better correlated with MRS (r2 = 0.904, P < 0.001) compared to IOP (r2 = 0.504, P < 0.001).

Figure 4.

FF map produced from the IDEAL images (see results) in a female subject aged 57 years indicating an increase in fat from L1 to L5. Color scale shows FF values in the range of 0% (dark blue) to 90% (bright yellow).

In six subjects BMD measurements were also obtained. Figure 5 plots a comparison of BMD against FF (determined using MRS) for these individuals. There is a negative trend between BMD and FF, which was of borderline statistical significance (P = 0.076). Of note, there is considerably more population variation (coefficient of variation [CV] = 100 × SD/mean) in MRS values, both individually and as a group, in comparison with the BMD values (CV equal to 13% and 31% for the BMD and MRS measurements).

Figure 5.

A plot of BMD vs. FF in six subjects. Each point is a single vertebral measurement and identical markers are used for data from the same individual (male: open markers, female: solid markers). There is a negative trend between BMD and FF but a much wider variation of FF values.

Two subjects had more than one MRS investigation as part of the study: In the first subject (female aged 57 years), repeat examinations were performed to assess the repeatability of the technique. Values of FF from these scans were found to be 68.9% and 67.9%. Figure 6 compares two spectra from the L3 VB in a male aged 34 years who was scanned at both 1.5 Tesla (Fig. 6a) and 3.0 Tesla (Fig. 6b) and processed in an identical manner. Although the signal-to-noise ratio is reduced in the second spectrum (acquired with half the number of signal averages), a comparison between the two clearly shows the improvement in spectral quality with smaller lipid resonances visible at the higher field strength (III and V) that are not distinguishable at 1.5 Tesla.

Figure 6.

A comparison of spectra from L3 in a male aged 34 years acquired at (a) 1.5 Tesla and (b) 3.0 Tesla.


Significant variation in normal bone marrow has been demonstrated using a clinical high field system. Age, gender, and skeletal site differences have been demonstrated and a large variance in FF was shown in subjects exhibiting similar measurements of BMD.

The established age-related increase in fat content has been investigated here for both an axial and peripheral skeletal site. In the lumbar spine there was a significant and gradual increase in fat relative to water with respect to age. When all L3 data points are considered (N = 15), the correlation is equivalent to a 7.5% increase in FF per decade. This compares favorably with previously reported values of 6.0% (11) and 7.0% (9). Results in the heel suggest a much earlier onset in the peripheral skeleton so that even at the age of the youngest subject studied here, there appears to have already been a substantial increase in yellow marrow. This concurs with findings from other nonquantitative MRI studies.

In order to convert peak ratios to absolute concentrations, or compare these results with other MRS published data, it is necessary to adjust for saturation effects. Using the sequence timings here (TR ≥ 2 seconds) there is little T1 relaxation and the major contribution is due to T2 relaxation, which changes little with field strength. Correction factors to consider for the PRESS sequence follow (1 – exp–TR/T1) and (exp–TE/T2) for the T1 and T2 relaxation, respectively. Previously published values of 901 msec/40 msec and 266 msec/73 msec for T1/T2 values of marrow water and fat can be found in Ref.19. These T2 values are similar to those quoted by Kugel et al (11) who also found them to be independent of age. Applying appropriate corrections to both this data and data given in Ref.10, which used a TE equal to 20 msec, reasonable comparisons can be reached. For example Schellinger et al (10) obtained mean FWR values in the age group 30–39 years of 1.14 and 0.53 for males and females, respectively (note: this data was at L2). These are slightly lower than our values of 1.25 and 0.55 for L3 data.

Results presented here add to the body of data that support gender variation in marrow fat content. To date, there has been a relatively equal number of MRI studies both for and against this finding. It is interesting to note, however, that only this present work and two other studies (10, 11) have used spectroscopy to examine this potential variation. Studies that have not observed any differences have all been based on imaging techniques. It is possible that those imaging techniques used for marrow quantification to either directly or indirectly measure fat content were less sensitive to these differences.

Our results also indicate a substantial variation in fat content within individual vertebral bodies of the lumbar spine. Although only 10 of the subjects were examined at more than one VB, there was a general increase in fat content from L1 to L5. The pathophysiological relevance of this is unknown, it may simply be reflecting the “peripheral to axial” conversion from red to yellow marrow with increasing age; since L5 is more inferior than L1, it could be expected that it would have a higher FF.

We believe that this is the first time bone marrow content has been examined in vivo at high clinical field strength. The advantages of higher magnetic field are an increase in both signal-to-noise ratio and chemical shift dispersion. The former may be translated into improved spatial resolution, which is crucial in the examination of small vertebral bodies and those that are anatomically more difficult to prescribe (e.g., L5), or reduced scan times, enabling the examination of multiple VBs. An increase in chemical shift dispersion is of benefit as it increases the separation of metabolite peaks and improves the potential for assessing the role of other fat resonances. One previous work looked at the presence of other fat peaks in osteoporotic patients and suggested there is a discrimination between saturated and non-saturated fat levels (20) with an increase in unsaturation (5.2 ppm resonance) as FF decreases. The measurement, however, may be confounded by either large water or fat resonances and the utilization of high field may prove useful in examining this relationship further. We have demonstrated an improvement in spectral quality at 3.0 Tesla compared to 1.5 Tesla in one subject. However, additional data is required to prove the benefits of the higher field.

Fat–water content is most accurately assessed using spectroscopy. However, quantitative imaging offers an improvement in scan time and resolution. We have examined two such techniques and found that both correlated with MRS values, with the IDEAL sequence being superior. This utilizes a more sophisticated in-phase and out-of-phase technique based on a three-point method and field map correction (18), and we believe this to be the first time it has been used in a quantitative manner. In the future, we plan to compare the accuracy of MRS and imaging techniques using homogeneous water–oil phantoms. The resolution of the MRS may be improved using chemical shift imaging and indeed this has been used to examine regional variation within the same VB (10). However, for investigations of more than one VB, we have found that the multiple single-voxel prescription described here is the only practical way of tailoring each voxel to lie completely within bone.

Recent studies have begun to link bone quantity with bone quality. Shih et al (21) showed a correlation between BMD and marrow perfusion in an older group of female patients. The same authors subsequently showed a link between bone marrow perfusion and lipid water content in female subjects (22). All three parameters were significantly correlated in a study by Griffith et al (23) examining the lumbar spine of elderly men. Their work showed that decreasing marrow perfusion and increasing fat content accompany a reduction in bone density. However, it could not be concluded whether or not there was a temporal relationship between these measures. We have begun to also examine the connection between BMD and FF and shown that MRS indicates a much greater variation in fat content over a comparatively smaller range of BMD values. However, further data is needed before conclusions can be reached.

In conclusion, these results contribute to the understanding of marrow content and variation in normal subjects. FF measurements may be added to a conventional MRI examination with only a modest increase in time. The technique could compliment other imaging techniques to provide a complete analysis of bone quality and integrity. Such examinations may have an important role not only in osteoporosis but also in studies of the late effects of cancer treatment.