Muscular lipids are involved in metabolic disorders (1, 2). Insulin resistance in particular, as an important predisposing factor for type 2 diabetes mellitus, is correlated with a significantly increased amount of myocellular lipids, even before measurable changes in the blood glucose level occur (3, 4). In addition, the lipid content in skeletal muscle may be affected by gender and age (5), training conditions (6), and recently performed exercises (7). All human skeletal muscles are composed of several different muscle fiber types. Up to seven different fiber types have been identified histochemically based on the pH stability of myofibrillar adenosine triphosphatase and on the myosin heavy chain profile (8). Innumerable fiber type transients exist due to continual adaptation processes. Slow-twitch fibers (type I with predominantly oxidative metabolism) show about five times more triglycerid content in biopsies than do fast-twitch fibers (types IIa and b with predominantly glycolytic metabolism) (7). For example, the soleus muscle mainly contains type I fibers, whereas the tibialis anterior muscle is the prototype of a glycolytic muscle.
Noninvasive and precise assessment of lipids in skeletal muscle could help to overcome problems inherent in invasive bioptic examinations. Quantitative X-ray computed tomography (QCT) reveals muscular dystrophy and myopathy with a replacement of the muscle bulk with fat and fibrous tissue (9). However, the low lipid concentrations under physiological conditions cannot be accurately quantified by QCT.
Volume localized proton MR spectroscopy (MRS) provides high sensitivity to muscular lipids (10). Two components of the muscular lipid signals are distinguishable because of their frequency difference of about 12 Hz at a field strength of 1.5 Tesla (11). It has been demonstrated by cases with fatty degeneration of musculature (11) and eosinophilia-myalgia syndrome (12) that one signal component stems from intramyocellular lipids (IMCLs), whereas the second component is related to extramyocellular fat (EMCL) nestled along the muscle fiber bundles. Further reports confirmed this thesis and provided insight into the spatial distribution of the lipid components by proton spectroscopy with variable voxel size and variable orientation of the muscle fibers (13). Unfortunately, the spatial distribution of lipids in musculature cannot be assessed adequately by spectroscopic techniques. Furthermore, postprocessing and interpretation of the spectral signals are more complicated than the visual analysis of usual MR images in routine clinical examinations.
Standard T1-weighted MRI provides enhanced visualization of lipid signals and shows fatty septa in the musculature and the surrounding subcutaneous fat, but small amounts of more homogeneously distributed fat in IMCL droplets with clearly less than 3% volume share cannot be assessed by usual T1-weighted techniques.
The present study reports on imaging sequences with improved selectivity and sensitivity to lipid signals compared with standard techniques. Parameters were optimized for a high signal-to-noise ratio (SNR) per measuring time and few artifacts in fat-selective MRI of the limbs using commercial volume coils. The developed imaging techniques were validated using phantoms containing lipid/water suspensions with a broad range of lipid concentrations. In a prospective study with 30 healthy volunteers, muscular fat imaging of the lower leg was compared with parallel localized spectroscopy in two muscle groups.
Feasibility of Muscular Fat Imaging
Biopsies from the human vastus lateralis muscle revealed an IMCL content of 0.13–0.17% in fast glycolytic and 0.38–0.99% in slow oxidative muscle fibers (14). Former spectroscopic studies in musculature of the human calf showed the integral signal (IMCL + EMCL) from methylene groups in skeletal muscle in a range of approximately 0.5% (m. tibialis anterior) to 10% (m. soleus), compared with the water signal integral for an echo time (TE) = 50 ms and a repetition time (TR) = 2 s (11). For voxel sizes of several cubic centimeters, the IMCL and EMCL portions in the spectra from soleus muscle were found to be of similar magnitude. SNR of musculature in standard fast spin echo images recorded with corresponding parameters (TE = 50 ms, TR = 2 s) provide an SNR > 300 for pure fatty tissue in an acceptable measuring time of a few minutes. For this reason, the signal component of muscular fat is expected to be slightly above the noise level, and the sensitivity of MRI to muscular fat should be sufficient using optimized parameters (TR, TE, and flip angles) and low receiver bandwidth. On the other hand, selectivity of the technique to fat must be clearly higher than in routine saturation techniques, since less than 1% of IMCLs must be identified in the presence of more than 50% water in musculature. The following section reports on the potential of spatial-spectral excitation techniques for this purpose. A further important aspect is that the size of the picture elements must be small enough to avoid artifacts from incorrectly encoded structures with high fat content (such as subcutaneous fat and bone marrow).
The new sequences were implemented on a 1.5 Tesla whole-body imager (Magnetom Vision; Siemens, Erlangen, Germany). All reported characteristics of lipid signals and sequence parameters are valid for the used field strength. However, adaptations to other field strengths appear to be possible.
Sequence Optimization for Muscular Fat Imaging
Spoiled gradient-echo sequences were chosen for fat-selective imaging of the musculature. A relatively high SNR per measuring time and a low general susceptibility to artifacts are inherent advantages of this technique.
Optimal measuring parameters depend on the relaxation characteristics of lipid signals, mainly from the methylene groups in the fatty acids (CH2)n. Former spectroscopic examinations (e.g., Refs. 10 and 11) revealed relaxation times of the dominating methylene signals of T1 ≈ 300 ms and T2 ≈ 90 ms. The TR in the fat-selective imaging sequence was chosen to 200 ms, since a further reduction of TR could only slightly increase the maximum SNR per measuring time. On the other hand, TR = 200 ms allows recording of several gradient echoes after each excitation for additionalT studies. For a given TR, the optimum signal yield can be obtained using Ernst's angle αE (15) for excitation, if transverse magnetization is spoiled after signal recording. In our case the optimum excitation angle was calculated to approximately 60°.
Selective imaging of muscular lipids requires high selectivity to the chemical shift range of methylene and methyl groups in fatty acids (0.8–2.0 ppm). Undesired excitation of protons with different chemical shifts (especially water protons at 4.8 ppm) must be clearly below the noise level. Previous work demonstrated the superior selectivity of frequency-selective excitation techniques (16, 17) compared to usual inversion recovery or saturation techniques. Several approaches with high selectivity in the frequency domain (including optimized Shinnar-Le Roux pulses) have been reported in the literature, but for slice-selective imaging an additional slice-selective 180° refocusing pulse would be necessary. Additionally, multislice imaging cannot be performed using these excitation techniques acting on the entire volume. Strategies for spatial-spectral excitation (18, 19) combine chemical shift selectivity (or Larmor frequency selectivity) with simultaneous slice-selective excitation and allow gradient-echo imaging. The recently reported excitation technique using equidistant slice-selective RF-pulses with (nearly) binomial amplitude ratios combined with refocusing field gradients between the single pulses (19) can be easily implemented on usual scanners, and allows off-center slice positions by conventional calculation of the pulse envelopes. For this type of spatial-spectral excitation an increasing number of RF-pulses in the binomial train provides narrower excitation bands besides broader frequency ranges without excitation (19). The number of RF pulses and their amplitudes were optimized considering the reported demands of fat-selective imaging. Figure 1a shows the pulse train in the final sequence with six pulses with amplitudes of 2°-(−9°)-17°-(−17°)-9°-(−2°) and a period of 2.32 ms. The parallel gradient switching provided slice selectivity and phase refocusing between the RF pulses. For this reason, no additional 180° refocusing pulse was necessary and the excitation flip angle of lipids could be adjusted to Ernst's angle. However, the excitation profile along the chemical shift and slice axis was recorded using a special sequence described in detail in Ref. 20. A homogeneous spherical sample (20 cm in diameter) filled with 5 mM NiSO4-doped water served as a phantom for the measurement of the excitation characteristics along the horizontal Larmor frequency axis and the vertical slice axis in Fig. 1b and c. The image in Fig. 1b shows the reported maxima and minima of excitation along the chemical shift axis with a period of 1/2.32 ms = 430 Hz. The transmitter frequency is adjusted to the Larmor frequency of water, resulting in maximum excitation at 1.4 ppm and minimum excitation at 4.8 ppm. The frequency range with less than 0.1% of maximum excitation has a width of 120 Hz and was found to be sufficient for achieving a complete lack of water signals from the leg after appropriate shimming. Although SNR > 500 in the maxima in Fig. 1b, a modified grayscale demonstrates the signal intensity (SI) in the non-excitation bands clearly below the noise level in Fig. 1c. A relatively low receiver bandwidth of 78 Hz per pixel was chosen, resulting in a readout period of 12.8 ms. Effects of transverse relaxation and of slight magnetic field inhomogeneities on spatial encoding remained uncritical using the mentioned bandwidth, and a series of six gradient-recalled echoes with TE = 16, 36, 56, 76, 96, and 116 ms could be recorded after each excitation. The sequence with TR = 200 ms allowed the recording of one slice with a thickness of 10 mm.
In cross-sections of the human calf, tissue with nearly 100% fat content is situated close to the musculature under investigation. In particular, the yellow bone marrow of the tibia and fibula, and subcutaneous fat consist nearly exclusively of lipids. Thus, for a precise determination of muscular lipids with clearly lower content (often < 1%), spatial encoding by the imaging sequence must lead to a correct representation of the tissue in the images. Figure 2 shows images from a bottle of sunflower oil recorded by a standard gradient-echo sequence. It is demonstrated that a Hanning raw data filter dramatically reduces artificial signals in the final image. But Hanning filtering does not provide suppression of all artifacts. Even such subtle manipulations as a person walking around the magnet led to additional contaminations of about 0.5% of the maximum intensity in regions beside the phantom in the phase-encode direction (Fig. 2d). Slight movements of the phantom, or slight instabilities of the RF amplitude of the transmitter might result in noticeable disturbances of phase encoding. This problem was found to be most critical in the development of highly selective imaging of muscular fat.
The signal dependence on the lipid concentration was assessed using a phantom with five parallel tubes (length = 100 mm, diameter = 25 mm). All tubes were completely filled with fluid and positioned parallel to the static magnetic field. The transmitter/receiver coil and the imaging parameters were chosen as in the volunteer studies reported below. One of the tubes was filled with pure water, and one with pure lipids (consisting of fatty acids with a chain length of 10 carbons and an iodine value of 94). The three other tubes were filled with lipid/water suspensions prepared with 0.1%, 1.0%, and 10.0% volume fraction of lipids (Fig. 3a). The fat-selective image from the entire sample in Fig. 3b shows sufficient selectivity and sensitivity of the proposed technique. The mean SIs of the tubes assessed in Fig. 3b were 0.18 (noise outside tubes); 0.20 (0.0% lipids, pure water); 0.36 (0.1% lipids); 1.4 (1.0% lipids); and 9.2 (10% lipids). All values were referred to the SI value of 100 for the sample with 100% lipids. The technique provides a nearly linear signal dependence on the lipid concentration. For measurements with lipid concentrations below approximately 1%, noise correction appears to be essential. The arrow in Fig. 3b indicates the problem of, with smearing signals in the surroundings of the 100% lipids tube due to insufficient phase encoding.
Comparison With Spectroscopic Examinations
Imaging of muscular lipids by the proposed technique in the human lower leg was compared to spectra from a representative region in the soleus muscle and in the tibialis anterior muscle in the same subjects (Fig. 4).
Analysis of the lipid content in the images was performed by the mean SI in the regions of interest (ROIs) corresponding with the cross-section of the desired muscle group. The borders of the ROI were chosen to be about 5 mm inside the intermuscular septa in order to avoid undesired contributions from fatty material in these septa. The fat content in the musculature was assessed using yellow bone marrow as reference with 100% fat content. The resulting percentage fat content in the musculature was calculated, including noise correction, by
This correction was applied on the final magnitude images. Gaussian distribution of noise in the complex image data is presupposed, but not necessarily fulfilled in the measurements. However, the corrected signal amplitudes clearly are more reliable than uncorrected data.
The spatial inhomogeneity of the RF field characteristics of the extremity coil was found to be negligible (< 2%) over the cross-section of the leg, and no correction was performed in this field.
Spectra were recorded from representative regions of the soleus and tibialis anterior muscle by a single-voxel stimulated echo acquisition mode (STEAM) technique (21) with TR = 2 s, TE = 10 ms, and TM = 15 ms from volume elements of 12 mm × 12 mm × 20 mm. Volume selection was performed by numerically optimized RF pulses (provided by the standard software of the manufacturer) with a duration of 2.56 ms in the presence of 6.0 mT/m field gradients. Averaged spectra from 40 scans recorded without water suppression were analyzed. The percentage fat content was assessed using integral SIs of methylene and methyl components in the spectra (integration range 0.8–2.0 ppm). The integral of the water resonance (integration range 4.1–5.4 ppm) served as reference for the assessment of muscular lipids. This procedure appears to be feasible, since fat-suppressed images from musculature indicate a homogeneous distribution of water in the entire cross-section of the lower leg. The SNR level in the spectra was much higher than in imaging examinations, and noise correction was not necessary. The quantification of muscular fat by spectroscopy was performed by
It should be noted that the percentage data from the described procedure with water as internal reference could differ from data using pure fat in the subcutaneous layer as internal reference. For this reason, absolute values of “muscular fatimaging” and “muscular fatspectroscopy” cannot be directly compared, but the statistical correlation of the data reveals whether both techniques are sensitive to the same property of tissue.
A prospective study on the calf musculature of 30 healthy volunteers (16 males and 14 females) was performed. The subjects (Table 1) underwent examinations in a 1.5 Tesla whole-body unit in the supine position. After positioning the lower leg in the extremity coil (a standard transmit/receive circularly polarized volume coil), images and spectra were recorded from regions about 10 cm below the knee joint.
Table 1. Age, Anthropometric Data, and Ranges of Findings in MR Fat Selective Imaging and Localized Spectroscopy in 30 Volunteers
Males (N = 16)
Females (N = 14)
BMI, Body mass index (kg body weight/m2 body surface).
Mean ± SD
29.4 ± 7.3 yr
31.4 ± 8.7 yr
Mean ± SD
23.9 ± 3.3 kg/m2
23.3 ± 4.6 kg/m2
Lipid content (Tibialis ant.)
Mean ± SD
2.0 ± 0.7%
3.1 ± 1.2%
Mean ± SD
Lipid content (soleus)
Mean ± SD
3.1 ± 1.1%
3.3 ± 1.5%
Mean ± SD
3.3 ± 1.1%
3.5 ± 1.5%
In addition to MRI and spectroscopic examinations, anthropometric data (body mass index (BMI), relative amount of subcutaneous fat in the cross-section of the lower leg at maximal diameter) of all subjects were assessed and reported in Table 1. Correlations between results from fat-selective imaging (muscular fatimaging) and from corresponding volume localized spectroscopy (muscular fatspectroscopy) were tested for two muscle groups (soleus and tibialis anterior). In addition, potential influences of age and BMI on the muscular lipids were investigated.
The proposed technique for fat-selective imaging provided images with sufficient quality in all subjects. None of the images showed problems with selectivity to lipids, and the non-excitation range of 120 Hz was found to be broad enough even in the expected presence of slight field inhomogeneities. Especially the marrow fat of the tibia showed problems with signal smearing in the phase-encode direction. This finding may have been partly caused by pulsatile movements induced by the popliteal artery. A similar effect was detected for subcutaneous fat, but it was less pronounced (e.g., Fig. 5c).
The distribution of lipid signals in the lower leg showed interindividual differences, as demonstrated by three examples in Fig. 5. There was a tendency to relatively high lipid content in the peroneus muscle groups and in the soleus muscle. SIs in these regions were clearly above the noise level. In contrast, the tibialis muscle groups appeared only slightly brighter than the background noise, and the number and extent of fatty septa between the muscle bundles were also very low in this region.
Both fat-selective imaging and spectroscopy revealed marked interindividual variability of the lipid content in the soleus and the tibialis anterior muscle. Table 1 shows the minimum, maximum, and mean values in male and female volunteers, with differences in the total fat up to sixfold in the same muscle group. The correlation between the results from fat-selective imaging and localized spectroscopy are presented in Fig. 6. For the tibialis anterior muscle the percentage fat content derived from fat-selective imaging was higher than the spectroscopic results in all but one of the volunteers (Fig. 6a); this was not consistent for both techniques, resulting in a low correlation coefficient of r = 0.55. Signal smearing from adjacent subcutaneous fat seems to be the main factor for the overestimation of the fat content in the tibialis anterior muscle by the imaging approach (see examples in Fig. 5, and also signal smearing in the phase-encode direction in Fig. 3b). In addition, rarified fatty septa in the tibialis anterior muscle were avoided in the selected volume elements for spectroscopy, but partly included in the selected ROIs in the images (see Fig. 4). In contrast, the soleus muscle provided high correlation (r = 0.91) between imaging and spectroscopic values. The conditions in the soleus muscle (markedly higher fat content (about 4%), more regular intermuscular fatty septa, and lack of adjacent fat layers) provided more accurate values of the proposed fat-imaging technique for this site.
Although the selected volunteers did not provide a representative demoscopic study group, the dependence of the muscular fat content on age and on BMI was tested. Only a slight increase of the fat content in the soleus muscle was found for increasing age of the volunteers. In addition, volunteers with a BMI of less than 30 kg/m2 did not show a significant correlation between the amount of lipids in the soleus muscle and BMI. However, three obese volunteers with BMI > 30 kg/m2 had a very high muscular fat content of up to 8%. This finding indicates that marked obesity can be associated with a modified muscular lipid distribution.
The proposed fat-selective imaging technique allows assessment of muscular lipids in the human calf, especially in regions remote to subcutaneous and bone marrow fat. Measurements in homogeneous lipid/water suspensions indicated nearly linear signal dependence on the lipid concentration. However, the gradient-echo approach is expected to be sensitive to microscopic susceptibility effects leading to undesired signal losses, especially for longer TEs. Insufficient spatial encoding can result in smearing of signals from tissue with very high fat content in the phase-encode direction, and to incorrect SIs in adjacent musculature with low lipid content. Consequently, the assessment of lipids in the tibialis anterior muscle was problematic, but most other muscle groups are expected to be less influenced by these contaminations. The phenomenon of relevant signal smearing was shown to be present in standard 2D MRI and to be not at all specific for fat-selective imaging. In phantom measurements (Fig. 3) the contaminations with artificial signals in the surroundings of the sample with 100% lipids were clearly more intense than the 0.1% lipid sample itself. The assessment of 0.1% lipids in tissue close to pure fat requires a very precise spatial encoding. Modified spatial encoding strategies with spiral pathways through k-space, or 3D approaches could help to overcome these problems.
Both applied strategies for the assessment of muscular fat—localized proton spectroscopy and the new fat-selective imaging technique—have inherent advantages and disadvantages. In contrast to localized proton spectroscopy, fat-selective imaging was not capable of distinguishing IMCL and EMCL portions. However, preliminary TE-dependent signal measurements revealed a different behavior of muscular fat compared to pure fatty tissue in the fat-selective images. This finding results from the broad Larmor frequency range in picture elements containing portions of both IMCL and EMCL. The frequency dispersion in the picture elements leads to signal dephasing and to a more rapid signal loss for longer TE in the musculature. The frequency distribution in the examined volume element can be directly assessed by spectroscopic examinations (11–13). Since direct measurements by means of fat-selective imaging are not possible, the signal phase and the signal decay could be the key for the desired selective IMCL imaging in future applications.
A new aspect provided by the fat-selective imaging technique is the assessment of the spatial distribution of lipids. The presented cases demonstrate a marked interindividual variability of the muscular lipid distribution. The examples show that fat-selective imaging has the potential to provide a new and interesting tool for studies in endocrinology and sports medicine. It is known that even short-term interventions, such as lipid infusions in the presence of hyperinsulinemia (22), and a moderate workload for several hours (13,23) result in a change of the IMCL level of about 30–50%.
Sensitivity to fat could be improved using smaller receiver coils (instead of a standard extremity coil of about 20 cm diameter) and new-generation MR imagers with field strengths higher than 1.5 Tesla. Optimized acquisition strategies with less sensitivity to artifacts and/or spatially selective saturation of subcutaneous and marrow fat could reduce the above-mentioned problems with signal smearing in the phase-encode direction.
For these reasons further effort should be made to develop and establish fat-selective imaging, which would enable clinical examinations in the large number of patients with pathological lipid metabolism.
The members of Siemens Medizintechnik are acknowledged for their continuous support.