Water and fat (triglyceride) are the two major signal sources in MRI. Fat signals can compromise MRI quality by overwhelming the water images in fat-abundant regions. A strong fat signal can also impede the precision of tumor detection by inducing chemical shift artifacts and by obscuring the effects of MRI relaxation agents, which interact only with water (1, 2). Therefore, fat-suppression and water- or fat-selective imaging have been used in many routine clinical examinations to reduce artifacts and to improve tissue characterization (3). Experimental techniques to achieve these goals include direct presaturation on the large fat methylene peak (4, 5), selective suppression of fat based on the T1 relaxation time difference between water and fat signal (6–8), selective excitation of water using selective pulses (9–11), and the ‘Dixon method’, which works with phase shifts (12, 13). These methods can be very sensitive to the heterogeneity of the amplitude of the static field and/or the radiofrequency (RF) field (3). Furthermore, the majority of these schemes consider fat to have only a single resonance or assume that all the fat signals have a uniform relaxation time. In reality, fat triglyceride comprises 10 resonances that are spread over a chemical shift range of more than 4.4 part per million (ppm) from the vinyl peak to the methyl peak. The strongest peak is the methylene on the triglyceride fatty acid chains, excluding those methylene groups that are one bond away from the vinyl group and those one or two bonds away from the carbonyl group. This large methylene peak accounts for approximately 60% of all the fat signals. The remaining 40% of fat signals have been ignored in most fat-suppression techniques; this can induce image artifacts and compromise image interpretation. The problem is typically more pronounced at higher magnetic field.
Here we show a new approach for fat suppression that completely eliminates all the fat signals irrespective of their chemical shift frequency and that is insensitive to the heterogeneity of RF field and static field. This new approach relies solely on the fact that fat protons do not exchange magnetization with water or with protein and cell membrane phospholipids protons in tissue samples (14). This method extends the classic MRI magnetization transfer (15–17) technique and generates a fat-free image that is sensitive to magnetization transfer, as well as the tissue water density and relaxation time.
MATERIALS AND METHODS
Approach for Elimination of the Fat Signal
Figure 1 illustrates our experimental approach. Figure 1a shows the basic magnetization transfer network in tissue for the three components relevant to our approach: tissue protein and membrane phospholipid, water, and fat. Magnetization transfer between fat and water, and between fat and the protons of tissue protein and membrane phospholipid, either does not exist or is extremely weak and not detectable with current NMR methods (14). In contrast, magnetization transfer between water and the tissue protein and membrane phospholipid proton component is complex but highly efficient. This magnetization transfer network suggests that presaturation of this tissue proton component will have no effect on the fat signal but will partially or fully saturate the water signal due to magnetization transfer from water to the tissue protein and membrane phospholipid protons. On static state NMR analysis, the resonance from tissue protein and membrane phospholipid protons has a line width that exceeds 2000 Hz (18, 19). The proton signal from this tissue component is thus too broad to be seen in the static NMR spectrum and does not contribute to the MRI signal. Selective presaturation of the protein and membrane phospholipid tissue protons is easy to implement; for example, by using a low-power, long-time irradiation at any position along the broad spectral resonance frequency of this tissue proton component (15, 16), as long as it is not on water or fat signals.
The method outlined in Fig. 1 is further described using a simple mathematical model. The signal of a regular image (schematized in Fig. 1a) includes both water and fat signals:
Here, sw and sf are the image signals from water and fat, respectively. This image can be acquired using any approach that is best suited to the particular application. Therefore, signals sw and sf can be sensitive to water density and fat and water relaxation times.
Figure 1b schematizes a second image acquired with the tissue protein and membrane phospholipid protons presaturated. This presaturation does not change any of the fat signals because there is no magnetization transfer from fat to water or to the tissue proton component. Magnetization transfer from water to the tissue protein and membrane phospholipid protons will reduce the water signal to αsw, where α describes the proportion of the water signal that remains. Consequently, the signal acquired in Fig. 1b consists of the full fat signal and a portion of the water signal:
Factor α depends on the efficiency of magnetization transfer between water and the tissue protein and membrane phospholipid protons and is determined mainly by the magnetization transfer rate between water and these tissue protons, the relaxation rates of water and these tissue protons, and the irradiation time and intensity of the pulse used to presaturate the tissue proton component (20–23). Factor α can be experimentally determined for a specific tissue.
A subtraction of the two images represented by Eq. 1 and Eq. 2 results in the final image schematized in Fig. 1c:
All the fat signals are eliminated irrespective of their chemical shift.
This method is independent of the heterogeneity of the static or RF field. In the water-only image that is proportional to (1 − α)sw, the magnetization transfer factor (1 − α) provides magnetization transfer contrast (15, 16) in addition to the contrast provided by water density or relaxation time embedded in the image signal sw.
If the factor α is estimated by an independent experiment, an image that consists only of fat signal can be determined from Eq. 1 and Eq. 2 as
This fat-only image contains signals from all fat resonances. If α approaches 0, this calculation is unnecessary; Fig. 1b is a fat-only image.
Tissue Sample Experiments
Experiments were carried out on a Bruker Avance 600 MHz spectrometer (Bruker Biospin, Billerica, MA) and on a 5mm probe with x, y, and z triple gradients. The samples were surgically resected human fat tissue and well-differentiated liposarcoma tissue. Specimens were obtained with consent of the patients and with institutional review board approval. Each tissue sample was cut into small pieces. The small pieces were then drawn into a capillary tube (inner diameter 1mm; outer diameter 1.8 mm) by attaching the capillary to a syringe. In aggregate, these small tissue pieces occupied about a 1-cm-long length of the capillary tube. One end of the capillary was then sealed with a small piece of polytetrafluoroethylene seal tape. A second capillary filled with heavy water (D2O) for field-frequency lock was placed parallel with the sample capillary into the 5mm tube. The samples were maintained at 20°C during experiments.
The fat-suppression method was applied to both spectra and images. The images were acquired using presaturation and gradient echo with presaturation power off and on, respectively. Detailed parameters for imaging are listed in the figure caption. To clearly illustrate the imaging results, a small field of view was used. For spectroscopy, a 90° pulse with presaturation was used. The spectra were acquired using eight scans, 16,000 data points. The spectrum was acquired on the whole tissue, and the thickness of the image slice was 5.5 mm.
For both imaging and spectroscopy, the RF frequency was at 8.3 ppm for presaturation and was moved to 4.7 ppm for pulsing and data acquisition. The presaturation pulse was 5 sec long, with an intensity of approximately 100 Hz.
Obese Mouse Imaging
The experiments were carried out on an obese mouse (model number OB-M-F, Taconic Farms Inc, Hudson, NY) under a protocol approved by Memorial Sloan-Kettering Cancer Center Research Animal Resource Center. Images were acquired on a 7.05-T Bruker Biospec Spectrometer using a birdcage coil. A coronal image was acquired with rapid acquisition with relaxation enhancement (RARE) (24) and presaturation. Imaging parameters included echo time = 35 ms, pulse repetition time = 2665 ms, field of view 4.2 × 5.6 cm, matrix 256 × 256, slice thickness of 1.06 mm, and total imaging time of 20 min.
The pulse used to selectively presaturate the tissue protein and membrane phospholipid protons consisted of 250 repetitions of (10 ms gaussian pulse–10 ms blank) with total length of 5 sec and an average intensity of approximately 110 Hz for the gaussian pulse. Two images were acquired, with the presaturation power off and on, to allow for a subtraction to obtain the fat-free image. As in the above tissue experiment, the RF frequency was at 8.3 ppm for presaturation and was moved to 4.7 ppm for pulsing and data acquisition.
Dispersion of Fat Resonances and the Resulting Chemical Shift Artifact
It is well known that fat signals induce chemical shift artifacts into a water image (1–3). To illustrate these artifacts, we performed spectroscopy and imaging without fat elimination on a sample of human fat tissue. Fig. 2 shows the spectrum (Fig. 2a) and the contour-plot image (Fig. 2b) acquired on a sample loaded in a capillary tube with a diameter of 1mm on a 600-MHz (14 T) spectrometer. The water signal from the fat tissue (peak 10 and a small fairly broad peak under peak 10) was very small compared to the strong fat signals. The major methylene peak (peak 2) accounted for 59% of all fat signals in this sample. The vinyl signal (peak 12) was approximately 11% the intensity of peak 2.
The broad separation of fat resonances results in a chemical shift artifact, in which the signals originating from low resonances are misregistered in the image with respect to signals from high resonances (1–3). At high field, the resonance frequencies of the different fat signals show increased separation, which can further intensify this chemical shift problem. Thus, in Fig. 2b, which is an image generated at high field strength (14 T) and with a small field of view, it is not surprising that the various fat signals were misregistered, resulting in two images. The left image came from resonances 7, 8, 11, and 12 of fat. The small tissue water signal (resonance 10) also contributed slightly to this left image. The right image came from resonances 1, 2, 3, 4, 5, and 6, with the majority of signal generated from the methylene peak 2. The intensity of the left image is approximately 8.5% that of the right image. The 1.1mm separation of the two images is consistent with the 2419-Hz distance between vinyl peak 12 and methylene peak 2 in the spectrum at 600-MHz field and with the imaging parameters (applied reading gradient of 1.1 gauss/cm; acquisition sampling rate 40,000 Hz). In addition, the spread of the fat resonances 1, 3, 4, 5, and 6 around 2 generated extended edges on both sides of the right image (see the areas in the dashed rectangles), as evidenced by its shape being elliptical rather than circular. If conventional fat suppression were applied to remove the signal from peak 2, both images would persist, as would the extended edges around the right image.
Complete Elimination of Fat Signal in MRI of a Liposarcoma Sample
The magnetization transfer methodology for complete fat elimination was applied to sample of well-differentiated liposarcoma, a tumor type that typically contains triglyceride at a concentration around 5 M (see Fig. 3a). In this specimen, the water signal was weaker than the strong fat signal but was large enough to be imaged. The spectra (Fig. 3a-c) illustrate the concept of complete fat-suppression. The spectrum in Fig. 3a was acquired without presaturation and that in Fig. 3b was acquired with 5 sec presaturation at 8.3 ppm. The spectrum in Fig. 3c, which is the subtraction of the spectrum in Fig. 3b from 3a, shows clean elimination of all the fat signals.
As with the fat sample in Fig. 2, the liposarcoma sample was misregistered into two images, both in imaging with presaturation (Fig. 3e) and without presaturation (Fig. 3d). However, unlike the fat sample, for which both images were from fat because the sample contained very little water (Fig. 2b), in this well-differentiated liposarcoma the water signal (peak 10) was stronger than the vinyl peak from fat. Thus, the left images in Fig. 3d and e were mainly from water, though the vinyl and the glycerol proton resonances from fat also contributed. The right images in Fig. 3d and e were purely from the fat resonance, mainly from the methylene at 1.29 ppm. The images show that water signal was not spatially homogeneously distributed inside the tube. It is unknown if this nonhomogeneous distribution reflects the true nature of the tissue water or if it resulted from the sample preparation process with two distinct small pieces of liposarcoma tissue residing in an imaging slice from the same axial location in the capillary tube. The image in Fig. 3f is the subtraction of Fig. 3e from Fig. 3d. In Fig. 3f the large image from fat was completely eliminated.
MRI of an Obese Mouse
Application of this method on an imaging scanner is straightforward. Fig. 4 shows images from an obese mouse. Fig. 4a is a regular image acquired using RARE (24) with no presaturation pulse. This image displays large quantities of fat all over the body of the obese mouse. The signal intensities of kidney and liver are much weaker than that of fat. Fig. 4b was acquired using the same method, but with a presaturation pulse. The kidney signal was barely detectable, implying that the 5-sec irradiation of the tissue protein and membrane phospholipid proton component resulted in very efficient magnetization transfer from water. Figure 4c is the subtraction of Fig. 4b from Fig. 4a. We increased image brightness/contrast of Fig. 4c compared to Fig. 4a and b. In this fat-free image, fat suppression was uniform and the image contained no imaging artifacts. For kidney, the signal intensity of the subtracted image (Fig. 4c) was 85% of the standard image (Fig. 4a). The 15% reduction in signal intensity results from the magnetization transfer factor α and removal of the small fat signal that originates from the low fat levels present in kidney. The signal intensity for liver in Fig. 4c was reduced more than 50% from the standard image because the liver of this obese mouse contained a substantial amount of fat that was no longer detected in the fat-subtracted image (Fig. 4c).
DISCUSSION AND CONCLUSION
We have shown that fat signal can be completely eliminated from MR images by exploiting the absence of magnetization transfer from fat to water or to protein and membrane phospholipid protons. The method entails imaging with and without presaturation of the tissue proton component to suppress signal from water, and subsequent generation of a water-only image by subtraction. We demonstrated clean elimination of fat signals for a liposarcoma sample and an obese mouse. Both tissue sample and the obese mouse contained concentrations of triglyceride that exceeded 5 M and therefore represent an extreme for the amount of fat relative to water. This demonstrates that the proposed fat-suppression method can be applied to eliminate the fat signals for tissues with any fat content.
The fat-elimination method described is substantially more efficient at suppressing fat signal than the conventional fat-suppression and water- or fat-selective imaging methods. The major methylene signal is the focus of most traditional techniques for fat-suppression, water-selective imaging or fat-selective imaging, while the remaining resonances are ignored. For a regular triglyceride molecule including two double bonds in each of the three chains, the major methylene signal constitutes at most 65% of all the fat signals. Any increase in the double bond content of fat will further reduce the ratio of fat methylene signal to total fat signals in the sample. Therefore, traditional fat-suppression techniques that address only the major methylene signal fail to suppress at least 35% of the signal from fat. In addition, presaturation (4, 5), which is among the most popular fat-suppression methods, will reduce the water signal intensity because saturation of the major methylene signal also partly saturates protein and membrane phospholipid signal, which triggers magnetization transfer from water.
Our fat-elimination approach offers several advantages. As shown in Fig. 4 of the obese mouse, we retained 85% of the kidney signal intensity. For most tissues other than fat, magnetization transfer from water to protein and membrane phospholipid protons is very efficient (α is very small), and thus in most tissues very little water signal is lost. Furthermore, the elimination of the fat signal does not depend on the chemical shift or the line width of the fat signals, and it is not affected by the heterogeneity of amplitude of static field or RF field. The magnetization exchange method can be utilized for any magnetic field strength, and an increase in magnetic field strength does not require any further modification to the approach. Finally, this approach can be combined with other MRI techniques.
The only drawback to this approach (which is shared by all fat-suppression methods that are based on the subtraction of two images) is the 2-fold increase in imaging time. However, the acquired water-only image provides additional information, as it is now sensitive to magnetization transfer in addition to the tissue water density or relaxation time. Magnetization transfer imaging is a well-developed methodology (16, 17, 25) and has been found useful in many applications such as magnetic resonance angiography (26), breast imaging (27), brain metabolism (28), and multiple sclerosis (29). The fat-free imaging method presented in this work retains all the advantages of traditional magnetization transfer contrast. It should further enhance contrast for more precise tissue characterization and tumor delineation, particularly for tumors arising in regions of high fat content such as the retroperitoneal space.
The magnetization transfer factor (1 − α) in Eq. 3 depends on many factors (20–23) but can be controlled by adjusting the intensity and length of the presaturation pulse. When the presaturation pulse is strong enough and long enough, (1 − α) are approximately the same for many tissue types (data are to be published in a future paper) and thus will not contribute contrast to the fat-free image. However, under standard imaging conditions, the intensity and length of the presaturation pulse are typically very limited, resulting in significantly different magnetization transfer rates for different tissue types and under these conditions will enable significant magnetization transfer contrast.
The authors thank Rachael O'Connor for her assistance with the mouse experiments and Dov P. Winkleman for imaging the obese mouse on the 7-T scanner. This research was supported in part by the Kristen Ann Carr Fund and the Department of Surgery at MSKCC, and the soft tissue sarcoma Program Project P01 CA047179, and NIH P30 CA08748 and U24CA83084.