MR diffusion-weighted imaging (DWI) has widely been used over the past decades to measure molecular water diffusion in vivo. DWI is a powerful noninvasive tool providing additional information about microscopic tissue compartments, structural anisotropy, and pathology of tissues (1–12). Additionally, diffusion properties of some metabolites with small molecular mass were shown to be accessible using diffusion-weighted spectroscopy (13).
MR imaging exclusively reflects signal contributions from water and fat molecules. In contrast to water-containing tissues, adipose tissue does not show any relevant signal attenuation using common b-values up to 2000 sec/mm2. The reason for the low sensitivity of adipose tissue to diffusion preparation in MRI is the difference in molecular size between water and fat: fatty compounds in adipose tissue consist mainly of triglyceride molecules with a molecular mass of ∼900–1000 u, compared to only 18 u for water. In triglycerides, glycerol is esterified with three fatty acids of different chain lengths and degrees of saturation. The composition and the amount of free fatty acids with clearly lower molecular weight than triglycerides are potentially interesting features of adipose tissue (14).
Besides temperature, the diffusion coefficient of fat molecules mainly depends on the molecular mass. In general, the diffusion coefficient of triglycerides is about two orders of magnitude smaller than the diffusion coefficient of water. For this reason, relatively high maximum b-values of ∼50,000 sec/mm2 are necessary to provide a diffusion-related signal attenuation of clearly more than 50%. Doing so, accurate assessment of the diffusion properties of adipose tissue seems feasible.
To ensure accurate assessment of diffusion coefficients in vivo, techniques should be insensitive to motion artifacts. Single-shot echo-planar imaging (EPI) is rather robust even under conditions involving slight coherent motion of the tissue under investigation and is therefore mainly used for DWI examinations. An alternative method for DWI is line scan diffusion imaging (15, 16), which was used for fat diffusion measurements in rat bone marrow by Ababneh et al. (17). Line scan diffusion imaging is also relatively insensitive to motion artifacts and even less sensitive to eddy current (EC) effects compared with EPI, but longer measurement time and limited signal-to-noise ratio have to be taken into account. In another study (18), a diffusion-weighted single-voxel spectroscopic stimulated-echo acquisition mode technique with b-values up to 80,000 sec/mm2 was applied to measure diffusion properties of aliphatic organic compounds of fatty tissue in human bone marrow and of intra- and extracellular lipids in skeletal muscle.
Extended body compartments with pure adipose tissue suitable for diffusion-sensitive MR examinations are (yellow) bone marrow of adults and subcutaneous and visceral adipose tissue. In this work, a diffusion-weighted twice-refocused spin-echo EPI sequence as introduced by Reese et al. (19) was modified and adapted to diffusion imaging of adipose tissue of the lower leg in a 1.5-T whole-body scanner. Additional studies in vitro on samples containing butanol and oleic acid were conducted to test the methodical approach.
Diffusion imaging of fatty components could play a role in the field of characterization of white and brown adipose tissue and in temperature measurements. On the other hand, diseased organ parenchyma with high amount of fat (as in liver steatosis or in musculature with fatty degeneration) could be probably characterized regarding their microscopic spatial and chemical composition.
MATERIALS AND METHODS
All experiments and examinations were performed on a 1.5-T whole-body MR unit (Magnetom Sonata, Siemens Healthcare, Erlangen, Germany). For signal detection, a transmit/receive extremity coil of the manufacturer was used.
On a whole-body scanner, long-lasting diffusion gradient pulses with high amplitudes are necessary to achieve b-values up to 50,000 sec/mm2. Those gradient pulses induce distinct ECs even in systems with actively shielded gradient coils. Unfortunately, because of the low readout bandwidth of EPI sequences in phase-encoding direction, those ECs might cause significant spatial distortions in the images, dependent on the direction of the applied diffusion gradients. To minimize such undesired image distortion effects, a twice-refocused spin-echo EPI sequence was applied as introduced in Ref. 19. In this sequence, spin-echo diffusion sensitizing is performed by two 180° refocusing pulses and four diffusion-sensitizing gradients with optimized durations: by appropriate adjustment of the timing of these diffusion-sensitizing gradients, ECs with a certain exponential decay constant can be nulled, and ECs with similar decay constants can be significantly reduced. For water diffusion measurements with lower b-values, this approach has been shown to generate clearly less distortion in diffusion-weighted images improving imaging quality of derived diffusion maps (19).
To avoid spatially shifted and disturbing signals caused by double bonds of fat molecules (and of water for lower b-values as present for in vivo measurements), all studies were performed using spectrally selective excitation. For this purpose, a series of six equidistant sinc pulses, separated by a delay of 2.4 msec, with an alternating binomial distribution of the flip angles and with monopolar switching of the slice gradients were used for spatial-spectral excitation in the chemical shift range of methylene and methyl protons of fatty acids. The field of view was chosen sufficiently large to avoid intersections of disturbing signals caused by N/2 ghosts with the basic signal in all experiments.
For all measurements, partial Fourier technique in phase-encoding direction was applied with a factor of 6/8 to decrease minimal echo time (TE). Before averaging and Fourier transformation in phase direction, linear-phase changes were automatically corrected in each raw data set by the customer's image reconstruction software using additional phase correction scans.
In Vitro Studies
Diffusion imaging was performed in two liquid organic compounds with different molecular weight filled in plastic bottles with a content of 1 L and a diameter of 80 mm. Bottles were placed horizontally inside the extremity coil parallel to the scanner bore axis. The two compounds were oleic acid (CH3(CH2)7CHCH(CH2)7COOH) and 1-butanol (CH3(CH2)3OH) with chain lengths of 18 and 4, respectively, and molecular weight of 282 and 74 u, respectively. For all measurements, temperature was in the range between 22.2 and 22.5°C.
For oleic acid, the sequence parameters used were according to the in vivo examinations as described below: pulse repetition time = 1500 msec, TE = 240 msec, matrix 128 × 128, field of view 256 × 256 mm2, receiver bandwidth = 1502 Hz/pixel, and eight averages. Data were acquired with 16 equally spaced b-values from 0 to 50,000 sec/mm2, with a maximal diffusion gradient strength of 32.2 mT/m. As a twice-refocused sequence (19) was used for diffusion weighting (instead of a common Stejskal-Tanner sequence), the b-value depends on four different time elements δ1, δ2, δ3, and δ4, rather than the usual diffusion times δ and Δ. For a TE of 240 msec, following time elements were applied: δ1 = 52.66 msec, δ2 = 51.18 msec, δ3 = 68.82 msec, and δ4 = 35.02 msec. For this choice of timing, ECs with an exponential time constant of 100 msec and their effects can be nulled.
For 1-butanol with its lower molecular weight, the TE was decreased to 140 msec, and data were acquired with 16 equally spaced b-values from 0 to 5000 sec/mm2 with a maximal diffusion gradient strength of 29.9 mT/m. Resulting time elements for appropriate diffusion weighting were δ1 = 20.00 msec, δ2 = 33.84 msec, δ3 = 36.16 msec, and δ4 = 17.68 msec.
One transverse slice with 6-mm thickness was chosen for all in vitro measurements. Three experiments were performed, each of them with the direction of the diffusion-sensitizing gradients along one of the three orthogonal scanner axes. Total measurement time was 3:17 min for one diffusion direction.
For diffusion analysis, circular regions of interest (ROI) with a diameter of about 50 mm were centrally placed within the bottle, and signal intensity against b-value curves was extracted for every pixel inside the ROI.
In Vivo Studies
Six healthy male volunteers (mean age 36 ± 9 years, body mass index 24.6 ± 3.2 kg/m2) were scanned for in vivo diffusion measurements of adipose tissue in the lower leg. All volunteers gave written informed consent before the examinations.
As a first step, multiple axial T1-weighted images of the right lower leg (fast spin-echo sequence with pulse repetition time = 650 msec, TE =16 msec, 6-mm slice thickness, matrix size 256 × 256, field of view 180 × 180 mm2, receiver bandwidth = 250 Hz/pixel, echo train length 3, and axial orientation) were recorded. For in vivo diffusion imaging, one single slice with 6-mm thickness was chosen positioned at the level with maximum cross-sectional area of the calf. Further sequence parameters of the twice-refocused spin-echo EPI sequence were pulse repetition time = 1500 msec, TE = 240 msec, matrix 128 × 128, field of view 256 × 256 mm2, receiver bandwidth = 1502 Hz/pixel, and eight averages.
To enhance the accuracy of apparent diffusion coefficient (ADC) measurements, data were acquired with 64 instead of 16 equally spaced b-values (as chosen for the in vitro experiments) from 0 to 50,000 sec/mm2 with a maximal diffusion gradient strength of 32.1 mT/m. Directions of the diffusion gradients were strictly parallel and orthogonal along the main axes of the scanner. Total measurement time was 12:53 min for one diffusion direction.
A maximal b-value of 50,000 sec/mm2, which is approximately the inverse of the expected diffusion coefficient, was used for the in vivo experiments to obtain distinct signal attenuation for higher b-values. Evaluation of mean ADC values was performed in ROIs located within the tibial bone marrow, where pixels containing marginal vessels were excluded. Further ROIs were selected in areas of pure subcutaneous adipose tissue and analyzed appropriately.
Diffusion-weighted images were analyzed offline on a PC using home-made routines written with Matlab® (The Mathworks, Inc., Natick, MA). After signal averaging, noise correction was applied to the magnitude signal Sorig of each picture element with , where σ2 is the variance of the noise (20). For σ2 an unbiased estimator with minimum variance can be derived by , where 〈R2〉 is the spatial average of the squared magnitude data out of a region in the image without MR proton signals (21). To avoid artificial values in the ADC maps in regions outside the object under measurement (plastic bottle for the in vitro measurements and lower leg for the in vivo measurements), a binary filter was applied, which considered only pixels containing signal of the object. The binary filter was defined by a polygon, tracing the outer contours of the object. For the pixels outside the polygon, signal was set to zero. A pixel-wise fit with an exponential function was applied to calculate the ADC values for the three diffusion directions. For a better visualization, ADC maps were color encoded. Mean ADCs were calculated for the selected ROIs within bone marrow and subcutaneous adipose tissue.
Diffusion experiments performed in vitro in fluids consisting of a well-defined chemical compound revealed logarithmic signal attenuation for increasing b-values. Figure 1 shows semilogarithmic plots of the signal decay against the b-value from oleic acid and 1-butanol. Signals were taken from circular ROIs with a diameter of about 50 mm centrally placed within the plastic bottles filled with the two organic compounds, respectively. Solid lines show the monoexponential linear fits, which are in good agreement with the measured data. Diffusion weighting along the magnet axis yielded ADCs of (3.36 ± 0.09) × 10−5 mm2/sec and (42.0 ± 1.2) × 10−5 mm2/sec for oleic acid and 1-butanol, respectively. Diffusion weighting in perpendicular direction led to comparable results with ADCs of (3.46 ± 0.11) × 10−5 mm2/sec (horizontal) and (3.37 ± 0.10) × 10−5 mm2/sec (vertical) for oleic acid and (40.5 ± 1.2) × 10−5 mm2/sec (horizontal) and (41.3 ± 1.2) × 10−5 mm2/sec (vertical) for 1-butanol, respectively.
Transverse T1-weighted images and corresponding series of diffusion-weighted images were recorded in six volunteers. Figure 2 shows a transverse T1-weighted image of the lower leg (Fig. 2a) of a 43-year-old male volunteer (#2 in Table 1) and according ADC maps determined from the diffusion-weighted images with diffusion sensitizing along three orthogonal directions (Fig. 2b–d). Although ADC values of tibial bone marrow fat appear rather homogeneous, some regions in the subcutaneous adipose tissue reveal higher ADC values, partly depending on the direction of the diffusion gradients. Figure 3 shows semilogarithmic plots of the signal decay against the b-value from ROIs within yellow tibial bone marrow of the same volunteer for all three diffusion directions. Solid lines show the monoexponential fits to the data, yielding diffusion coefficients of (1.69 ± 0.11) × 10−5 mm2/sec, (1.86 ± 0.13) × 10−5 mm2/sec, and (1.81 ± 0.16) × 10−5 mm2/sec, respectively. For the diffusion directions perpendicular to the static magnetic field, b-value-dependent signal decays in bone marrow were nearly perfectly fitting monoexponential functions. Data with the diffusion-sensitizing gradient along the static field B0 were more scattered around the monoexponential fit, especially for higher b-values. The finding of more pronounced scattering of the data for diffusion weighting along the magnet axis could be observed for all volunteers. One possible explanation for this observation may be the fact that the table of the scanner has more play along the scanner axis than in perpendicular direction. Therefore, vibrations of the table during application of the diffusion gradients may lead to additional axial movements of the volunteer's trunk, which are transferred to the lower leg. With diffusion encoding along the z-direction, this may lead to a more pronounced scattering of data for higher b-values. However, these supplemental motions seem to be too weak to significantly increase the derived mean ADC values for diffusion sensitizing in z-direction in comparison to the perpendicular directions. Negligible diffusion anisotropy effects in bone marrow are indicated. Table 1 provides ADC values from tibial bone marrow of all volunteers for all diffusion directions.
Table 1. ADC Values of Tibial Yellow Bone Marrow for All Subjects in Units of 10−5 mm2/sec
Mean values and standard deviation are given for diffusion sensitizing along orthogonal directions for a representative ROI.
1.95 ± 0.21
2.12 ± 0.28
2.28 ± 0.27
2.12 ± 0.17
1.69 ± 0.11
1.86 ± 0.13
1.81 ± 0.16
1.79 ± 0.09
1.72 ± 0.23
2.01 ± 0.25
2.04 ± 0.23
1.92 ± 0.18
1.86 ± 0.12
1.98 ± 0.15
1.84 ± 0.13
1.89 ± 0.08
1.76 ± 0.20
1.87 ± 0.15
2.01 ± 0.18
1.88 ± 0.13
1.88 ± 0.19
1.91 ± 0.18
1.76 ± 0.11
1.85 ± 0.08
30 ± 9
1.81 ± 0.10
1.96 ± 0.10
1.96 ± 0.20
1.91 ± 0.15
In contrast to the ADC values from tibial bone marrow, diffusion coefficients in subcutaneous adipose tissue showed a more heterogeneous behavior. Diffusion coefficients in some subcutaneous adipose tissue areas were in the same range as in tibial bone marrow (Fig. 2). On the other hand, areas with distinctly higher ADC values up to 4.2 × 10−5 mm2/sec occurred. It should be mentioned that especially in those regions with increased ADC values data presented low signal intensity and pronounced scattering around the fit curve for higher b-values.
DISCUSSION AND CONCLUSIONS
As the ADCs of fatty acids and triglycerides are about two orders of magnitude smaller than the diffusion coefficient of water, long-lasting diffusion gradients have to be applied to achieve sufficient diffusion-related signal attenuation. For testing of the proposed methodical approach, substances with high molecular weight were used for calibration purposes. Similar procedures were conducted in the earlier studies of Ababneh et al. (17) and Lehnert et al. (18). ADC values obtained in our phantom studies for 1-butanol and oleic acid are in good agreement with the results in Refs. 17 and18. Slight differences between ADC values for 1-butanol and oleic acid reported by Ababneh et al. and Lehnert et al. and the results in this work can be explained by different temperatures of the fluids.
Diffusion maps revealing ADC values of fatty compartments in the lower leg show good quality and reasonable results. Fatty tibial bone marrow led to a mean ADC value of (1.91 ± 0.15) × 10−5 mm2/sec with low variation within the cohort of six subjects. No significant dependence on the orientation of diffusion-sensitizing gradients was found, indicating that there is no pronounced anisotropy in bone marrow tissue. Based on these consistent findings, the method is considered suitable for a precise determination of fat diffusion coefficients in vivo. Furthermore, the ADC values are very similar to the results of Ababneh et al., who measured fat diffusion coefficients in rat bone marrow.
As mentioned in the Results section, isolated regions in the subcutaneous adipose tissue show distinctly higher diffusion coefficients in comparison to the findings in tibial bone marrow. Assuming a rather homogeneous distribution of fat diffusion coefficients in both tibial bone marrow and subcutaneous adipose tissue, an accurate determination of ADC values in these regions may be hampered by not yet identified problems. One possible reason could be fibrous connective tissue or vessels causing pronounced field inhomogeneities and signal voids in their vicinity. Furthermore, phase shifts caused by incoherent motion could lead to additional signal attenuation for high b-values and therefore to overestimated ADC values. There are several possible sources for incoherent motion: First, pulsation in blood vessels located in or adjacent to the fat tissue. To constrain this effect, an additional diffusion measurement with electrocardiogram triggering was accomplished for one volunteer. Measurements without electrocardiogram triggering and with electrocardiogram triggering with diffusion sensitizing in the diastole led to rather similar ADC maps, in contrast to a recent work of Brandejski et al. concerning spectroscopic-based diffusion measurements of intramyocellular lipids (22). Second, vibrations or small movements of the table due to forces present during gradient switching could lead to incoherent motion of tissue under examination. In an additional experiment, those small table movements due to gradient switching were decoupled from the lower leg by hanging up the leg and fixing it on a hook, which was not affected by motion problems of the table. However, no changes in the derived ADC maps could be obtained in comparison to the conventional bedding of the lower leg in the extremity coil.
High b-values up to 50,000 sec/mm2 required very long TE of 240 msec for in vivo measurements. The maximal gradient strength on our 1.5-T MR system was restricted to 32.1 mT/m by the safety monitoring of the scanner. However, because of the relatively long transverse relaxation time T2 of methylene signals from adipose tissue of about 90 msec (18), signal-to-noise ratio was sufficient for an accurate calculation of ADC values. The relatively short longitudinal relaxation time T1 of methylene signals of ∼300 msec (18) allows a short pulse repetition time leading to a moderate total measurement time, despite several averages were recorded for each image in a series with 64 different b-values. If necessary, the total measurement can further be distinctly shortened to 6:29 min or 3:17 min by choosing only 32 or 16 b-values, respectively.
Restricted diffusion in the presence of impermeable barriers might lead to non-gaussian diffusion characteristics, especially if long-lasting diffusion gradients are applied. With a mean diffusion coefficient D = 2 × 10−5 mm2/sec for fat and a total diffusion time T = δ1 + δ2 + δ3 + δ4 = 207.68 msec, mean squared displacement 〈z2〉 = 2DT for a fat molecule in one direction is only about 3 μm, which is small compared with the mean size of fat vacuoles in adipocytes of white fat (size ∼ 100 μm). For this reason, it can be assumed that diffusion in white adipose tissue is not restricted, and the probability density of the displacements is gaussian. Usual equations for calculating the b-values as used in Reese et al. (19) were applicable, and results showed a linear reduction of signal intensity for increasing b-values on a logarithmic scale.
In contrast to the study of Ababneh et al., diffusion gradients were only applied in one orthogonal direction for each single measurement and not simultaneously along all three directions. Of course, using all three directions for diffusion weighting would lead to a distinctly shorter minimal TE of 170 msec for the same maximal b-value and therefore higher signal-to-noise ratio. However, simultaneous application of gradients in y- and z-direction induces additional cross-term fields GyGzyz/2B0 (23–25). For the twice-refocused diffusion-weighted spin-echo EPI sequence with long-lasting diffusion gradients and high amplitudes, these cross terms lead to a net zero-order gradient moment, causing an echo shift in k-space and a signal loss in the image, especially for slices with an offset from the isocenter (26). In our measurements, slice positions were chosen close to the isocenter, while an offset y0 of 80 mm was necessary in phase direction. This effect, together with the relatively long TE, could be the reason for distortions and signal losses in images with high b-values, when a tetrahedral-encoding scheme was applied in some test measurements.
In Ref. 17, it was postulated that diffusion preparation with stimulated echo instead of spin echo may be more adequate to achieve high b-values with sufficient signal-to-noise ratio by keeping TE as short as possible to limit T2 decay signal loss with clinical scanners. Indeed, it can be shown that for water diffusion in skeletal muscle, signal yield will be higher for a stimulated echo in comparison to the twice-refocused spin-echo technique used in this work (27). However, the optimal choice for diffusion preparation with minimum relaxation-dependent signal losses depends on the ratio of T1 to T2 of the examined tissue (28). While for water in musculoskeletal tissues with a high ratio T1/T2, stimulated echo is more advantageous, the situation for fat is distinctly different. Because of the relatively long T2 and the rather short T1 of main fat signals (T2 = 90 msec and T1 = 300 msec for the methylene (18) and T2 = 170 msec and T1 = 500 msec for the methyl signals (29)), spin echo preparation is superior even for b-values as high as 50,000 sec/mm2. In this case, optimized values for the stimulated echo would be TE = 170 msec and mixing time = 130 msec for the methylene signals and TE = 175 msec and mixing time = 115 msec for the methyl signals. When compared with the spin-echo preparation, optimal stimulated-echo preparation would lead to 30 and 40% less signal yield, respectively.
It may be worthwhile to examine diffusion of adipose tissue in other regions of the human body, e.g., of subcutaneous and/or visceral adipose tissue in abdomen or thorax. However, because of the more pronounced field inhomogeneities in these body regions in comparison to the extremities, single-shot EPI may not be the adequate tool for DWI and the line scan diffusion imaging technique (15–17) may be a suitable alternative in this case.
The presented experiments show that MR imaging of diffusion in adipose tissue in vivo is feasible in relatively short measuring time and one may ask, whether there is any motivation for investigations into diffusion properties of fatty tissue. However, several potential applications of those measurements show up at the horizon:
First, temperature measurements in fatty tissue: usual temperature measurement based on the temperature-dependent shift of the water resonance is not applicable on fatty tissue. On the other hand, size and molecular weight of triglycerides inside fatty tissue can be expected to be rather constant. So, temperature can be considered as main determinant of measured ADC values. Diffusion of lipids could be a potential indicator of temperature inside the body, e.g., during hyperthermia, if current problems with physiological movements or movements generated by the MR scanner are overcome.
Second, assessment of the share of triglycerides and fatty acids in adipose tissue: fatty acids and triglycerides play an important role in metabolism. It is well known that storage of fat (mainly as triglycerides in the large vacuoles of white adipocytes) and transportation of fat (mainly as fatty acids in blood and through the cytoplasm) are of interest for studies in subjects with diseases as type 2 diabetes or obesity. The molecular weight of free fatty acids is roughly one-third of the molecular weight of triglycerides, resulting in faster diffusion of fatty acids. In cases with very high lipolysis in adipose tissue, one might expect an increased fraction of free fatty acids inside the cells and potentially a second component visible in the b-value-dependent diffusion signal curve.
Third, differentiation of brown versus white adipose tissue: in recent publications (30–32), it was reported that the amount of brown fat in humans is higher than assumed before, and the amount (and/or activity) of brown fat is reduced in subjects with obesity. Therefore, the amount of brown adipose tissue may be of importance for normal human physiology and a target for the research on obesity and diabetes. The lipid vacuoles in (plurivacuolar) brown fat are clearly smaller than in (monovacuolar) white fat. Therefore, lipid diffusion could be different in the small vacuoles of brown cells compared with the clearly larger vacuoles in white cells.
Fourth, differentiation of microvesical versus macrovesical steatosis of the liver: lipids can be stored in liver parenchyma in vesicles of different size inside hepatocytes. In patients with steatosis, fat fraction can be clearly higher than 10% and diffusion measurement of lipids by MR could get feasible. Nowadays, only histology allows distinguishing macrovesicular and microvesicular steatosis. This distinction is clinically important. Fat-selective MRI or MRS can only determine the total fat fraction, but not the microscopic distribution. Movement of lipids in very small vesicles is expected to be more restricted, but the vesicles themselves might be able to move inside the cytoplasm of the hepatocytes. Measuring diffusion properties of those hepatic lipids, which is more demanding than in pure adipose tissue, could lead to clinically interesting results.
Fifth, mobility and spatial distribution of intramyocellular lipids: a further example for restricted diffusion probably occurs for intramyocellular lipids, which are settled as little droplets with a diameter of clearly less than 1 μm in close apposition to mitochondria in the cytoplasm of myocytes (33). Analysis of the size of the droplets and their mobility inside the cytoplasm is interesting, as their availability for oxidation seems to be individually variable. The volume share of lipids within myocytes is often less than 1%. Therefore, the low signal-to-noise ratio is really challenging and needs further methodical development to obtain increased sensitivity for reliable diffusion measurements.
All those possible applications of lipid diffusion measurements are still speculative, but have currently no suitable noninvasive “competitors” in terms of providing similar diagnostic information. The diagnostic information is usually only available using biopsies or by implantation of probes (e.g., for temperature measurements). Altogether, there are several areas of motivation for measuring diffusion of lipids in adipose tissues or in fatty compartments of organs as liver and musculature.