Whole-body 3D water/fat resolved continuously moving table imaging


  • Presented in part at the 14th Meeting of the ISMRM, Seattle, WA, USA, May, 2006.



To study the feasibility of three-dimensional (3D) whole-body, head-to-toe, water/fat resolved MRI, using continuously moving table imaging technology.

Materials and Methods

Experiments were performed on nine healthy volunteers, acquiring 3D whole-body head-to-toe data under continuous motion of the patient table. Two different approaches for water/fat separation have been studied. Results of a three-point chemical shift encoding and a spectral presaturation technique were compared with respect to image quality and performance. Furthermore, fast, low-resolution, whole-body water/fat imaging was performed in two minutes total scan time to derive patient-specific parameters such as the total water/fat ratio, the intraperitoneal/extraperitoneal fat ratio, and the body mass index (BMI).


Good water/fat separation with decent image quality was obtained in all cases. The three-point chemical shift encoding approach was found to be more efficient with respect to signal-to-noise ratio (SNR) and acquisition time.


Whole-body water/fat sensitive MRI using continuous table motion is feasible and could be of interest for clinical practice. Some improvements of the method are desirable. J. Magn. Reson. Imaging 2007;25:660–665. © 2007 Wiley-Liss, Inc.

THE AMOUNT AND THE DISTRIBUTION of body fat are important diagnostic parameters. They are linked to a number of obesity-related diseases, such as diabetes and those of the circulatory system (1, 2). A clear knowledge about the fat and its distribution may help to provide and improve diagnosis and therapy of those diseases and could help to better assess risks for the individual patient. In this context, MR-based measurements of the water and fat distributions offer a promising approach. MR is noninvasive, various water/fat separation methods are already available (3–5) and the spatial resolution can be adapted to match the particular needs of the examination.

Obesity-related diseases often exhibit a systemic character. Thus, and from a comprehensive diagnostic point of view, the water and fat distributions are of interest in the entire body, making whole-body head-to-toe scanning desirable. Consequently, the patient table must be moved during the scan, since the extended anatomy to be covered is much longer than the homogeneous region of the MR scanner. Such extended field-of view (FOV) imaging techniques can be subdivided into two main approaches. In the so-called “multi-station” approach, signal sampling is performed in a successive manner at different table positions while the table is at rest (6). This approach entails frequent acceleration and deceleration of the patient table, which is not optimal with respect to sampling efficiency and patient comfort. Therefore, MR data acquisition during continuous table motion (7–13) has received significant interest recently, although it makes scanning technically more challenging.

In the present work, chemical shift–based water/fat imaging techniques were studied in combination with an efficient three-dimensional (3D) head-to-toe continuously moving table sampling technique (13). In particular, two water/fat separation approaches were compared with respect to image quality and performance. Furthermore, a fast, low-resolution water/fat resolved head-to-toe scanning method was investigated and used to derive patient-specific parameters such as the total water/fat ratio, the intraperitoneal/extraperitoneal fat ratio, and the body mass index (BMI) with only a small investment of scan time.


Continuously Moving Table Signal Sampling

Cartesian 3D gradient echo (GE) sampling under continuous table motion is considered with the aim to perform water/fat scanning in a single run through the magnet. From the basic concepts available (7–13), the lateral readout method was chosen, which is described in detail in Ref.13. In this method, the frequency-encoding direction is oriented left–right, that is, perpendicular to the direction of table motion, z. The radiofrequency (RF) pulse of the 3D GE sequence excites a slab of length L along z to restrict the area from which signal is received. The sampled data are corrected for the table motion, Fourier-transformed along z after each completed loop of z-direction phase encodings and then appended in hybrid space (kx,ky,z) (c.f., Fig. 1). After all data have been acquired, Fourier transformation is performed in the x and y directions. The lateral readout method was chosen in this work because it allows efficient data acquisition when the length of L is chosen to be much smaller than the usual FOV to suppress effects from field inhomogeneity. However, the studied water/fat separation methods (see below) could as well be incorporated into other moving table methods.

Figure 1.

Schematic illustration of continuously moving table imaging. Head-to-toe signal sampling, performed during table motion at constant velocity, v, is split into the acquisition of individual blocks of 3D data (left). In each block, signal is sampled from a volume of length L, which is seamlessly aligned in hybrid space (kx,ky,z). In this illustration, only one k-space coordinate is shown. The changes of the Larmor frequency, induced by the body susceptibility, are plotted as a function of z (right) for one volunteer. The numbers near the curve indicate the frequency offsets used for all volunteers to compensate for off-resonance in the magnetization prepared water/fat separation method.

Water/Fat Separation

Two basic water/fat selective approaches were investigated, which are schematically shown in Fig. 2. In the first one (Fig. 2a), a frequency-selective RF pulse (110°) of Sinc-Gauss shape, followed by a spoiler gradient, was employed for presaturation. The RF pulse was tuned to suppress either the water or the fat signal. Anatomy-dependent frequency offsets (Fig. 1) were applied to the RF pulse to compensate for the large body-susceptibility induced changes of the Larmor frequency encountered in a head-to-toe scan. These offsets were determined in a preparatory study in a number of volunteers to derive an approximately patient-independent lookup table that can be mapped to the patient's position. This allowed adjusting the RF of the magnetization-preparation pulses between the acquisitions of successive 3D k-space data sets.

Figure 2.

MR sequences for water/fat resolved imaging. a: Magnetization prepared method. A special spectral partial inversion recovery preparation scheme (SPIR) is applied to acquire water or fat signals using a segmented 3D GE sequence. The spectral bandwidth of the SPIR pulse used in this study is larger than the water/fat chemical shift to increase robustness. The frequency offset as a function of z given in Fig. 1 is applied to the SPIR pulse. b: Three-point chemical shift encoding method. Three consecutive GEs are sampled at each gradient polarity, using an EPI-type scheme. Each of the two three-echo data sets is used for individual water/fat signal separation.

In the second approach, a chemical shift encoding technique was applied (Fig. 2b). To allow water/fat separation under locally varying off-resonance, at least three GEs have to be sampled at different echo times (5, 14). Using a suitable signal reconstruction method such as the “iterative decomposition of water and fat with echo asymmetric and least-squares estimation” (IDEAL) approach (14), the off-resonance can be treated as an additional parameter and corrected in retrospective manner. In this work, an echo planar imaging (EPI)-type readout gradient was used to efficiently sample all echoes for a given phase encoding step. To avoid EPI-related phase errors, six echoes were sampled instead of three, which would be the required minimum. Thus, three echoes were sampled at each gradient polarity, providing data for two separate three-point water/fat reconstructions, which were added to improve signal-to-noise ratio (SNR). The addition could be performed in magnitude mode without any corrections because the chemical shift was negligible due to the chosen high pixel bandwidth.


In-vivo experiments were performed on a 1.5 T whole-body scanner (Achieva; Philips Medical Systems, the Netherlands). The homogeneous body coil was used for RF transmission and signal reception. During data acquisition, the patient table was moved at constant velocity. System settings for the whole-body scans such as resonance frequency, transmit/receive gains, etc., were determined once at the start of each scan in the abdominal/pelvic region, where the maximum MR signal and coil loading were expected. Slight oversampling (±13%) was performed along the z-direction in all scans to suppress effects of nonideal slab selection and to allow sufficient time for the spin system to approach steady state. Head-to-toe scans were studied according to three different protocols (described below). Nine healthy volunteers (27–38 years, all male) were involved in this study. The institute approved the study and informed consent was obtained from all participants.

Protocol I

In the spectral presaturation method (Fig. 2a), a virtual FOV of 512 (left–right × 2000 (foot–head × 256 (anterior–posterior) mm3 and voxel size of 2.67 × 2.67 × 5.33 mm3 were used. With a slab length of L = 101.3 mm, the entire virtual FOV acquisition consisted of 20 3D subsets of data. The water/fat selective preparation block of duration 19 msec was followed by segments consisting of four dummy cycles and 12 acquisition cycles using low-high k-space sampling order (TR/TE = 5.2 msec/2.5 msec, α = 15°, pixel bandwidth = 500 Hz). No relaxation delay was employed between an acquisition segment and the subsequent water/fat specific magnetization preparation. To allow the acquisition of head-to-toe data for both water and fat in a single run, the water/fat selectivity of the magnetization preparation pulse was alternated between the acquisitions of subsequent 3D subsets of data (12, 15). The matched table velocity was 2.59 mm/second resulting in a total scan time of 12.9 minutes. During reconstruction, the water- and fat-only data were separated.

Protocol II

In the chemical shift encoding method (Fig. 2b), the same voxel size, FOV and the same value for L were used as in Protocol I. The six-echo EPI-type 3D sequence (TR/TEfirst = 8.3 msec/1.5 msec, α = 15°, pixel bandwidth = 2338 Hz) covered k-space in linear order with a chemical shift encoding increment of ΔTE = 1.63 msec between consecutive echoes of the same polarity. Signal sampling was performed during a constant read-gradient. The matched table velocity was 5.29 mm/second, resulting in a total scan time of 6.4 minutes. The resulting six 3D data sets, referring to different effective echo times, were subjected to two separate iterative water/fat reconstructions for the odd and even echoes, respectively, which were combined in magnitude mode.

Protocol III

In the low-resolution chemical shift encoding method, a fly-back EPI-type (16) 3D GE sequence was used. After each RF excitation, three identically phase-encoded GEs were acquired at constant readout gradient of the same polarity. The sequence covered k-space in linear order with a chemical shift encoding increment of ΔTE = 1.22 msec (TR/TEfirst = 5.9 msec/1.3 msec, α = 10°, pixel bandwidth = 5446 Hz). The slab thickness L was set to 128 mm and the virtual FOV to 512 × 2000 × 294 mm3, covered with isotropic voxels of 6.4 × 6.4 × 6.4 mm3. The matched table velocity was 16.5 mm/second, resulting in a total scan time of two minutes.

After image reconstruction and iterative water/fat separation, additional data analysis was performed. Body height, weight (assuming a body fat density of 0.94 g/cm3) and BMI were estimated for each volunteer. Furthermore, the total-body water-to-fat ratio was determined and, after user-guided segmentation of the inner and the outer abdominal chamber, the ratio of intraperitoneal/extraperitoneal fat was determined as the ratio of fat inside/outside the abdominal chamber. All algorithms in this exploratory study were based on simple binary water/fat voxel counting, ignoring any partial volume effects. A voxel was considered to contribute to either the water or the fat count if its signal was above a 10% threshold of the maximum 3D signal.


All experiments were completed successfully and resulted in reliable image quality. In Fig. 3, selected slices of water/fat resolved 3D data are shown for one of the volunteers using protocols I and II. Good water/fat separation with comparable image quality is obtained with both approaches, although some deficiencies in image quality and artifacts are still present (see Discussion). A marked difference between the two protocols is that the chemical shift encoding approach is more time efficient, as it required only half the scan time to acquire water/fat data compared with the magnetization-prepared approach. Furthermore, its SNR is about 20% higher, mainly due to the 1.3-fold longer total sampling time of this EPI-type acquisition method.

Figure 3.

Head-to-toe water/fat resolved 3D imaging. One of 48 coronal planes (voxel size = 2.67 × 2.67 × 5.33 mm3) is shown for one volunteer, using the two different single-run approaches. a: 3D water image (left) and fat image (right) acquired using chemical shift selective magnetization preparation (total scan time 12.9 minutes). The arrows indicate the fat signal contamination in the water image and the signal drop.b: 3D water image (left) and fat image (right) acquired using the three-point chemical shift encoding method (total scan time 6.4 minutes). The arrow indicates the banding artifact.

Representative results of the low-resolution 3D water/fat scans (protocol III) are shown in Fig. 4. Excellent water/fat separation is achieved with almost no artifacts and high SNR thanks to the large voxel size. Selected slices of these isotropic data sets are shown for two different volunteers. Included are graphs of the water and fat integrals over x,y as a function of z, which helps to judge the composition of different body tissues of the respective volunteers. The parameters extracted from the data of the individual volunteers are summarized in Table 1.

Figure 4.

Low-resolution head-to-toe water/fat resolved 3D imaging. Selected views (coronal and sagittal) of the isotropic 3D water/fat data sets of two volunteers are shown. The data were acquired with isotropic resolution (voxel size = 6.4 × 6.4 × 6.4 mm3) in a total measuring time of two minutes using the three-point chemical shift encoding method. The curves show the projections, as a function of z, for the water and fat signals. Please note the similarities of the water projections in these two volunteers, whereas their fat projections differ significantly due to their different water/fat compositions.

Table 1. Whole-Body Water/Fat–Related Parameters Extracted From the Low-Resolution 3D Data for the Individual Volunteers*
Volunteer #Water/fatPeritoneal fat (i/e)Weight (kg)Height (cm)BMI (MRI)
  • *

    The total-body water-to-fat ratio (water/fat) and the ratio between intraperitoneal and extraperitoneal fat (i/e), which represents the ratio of the abdominal fat inside/outside the abdominal chamber, were determined. Furthermore, the height, the weight, and the body-mass index (BMI) were estimated. The values given in parentheses were obtained using conventional measuring methods.

12.110.3981 (80)183 (184)24
21.710.5178 (74)180 (180)24
33.270.5370 (69)172 (172)24
42.300.5376 (74)180 (182)24
52.040.3680 (80)179 (180)25
61.630.31104 (97)194 (192)27
71.320.4183 (84)173 (174)28
81.170.4988 (85)174 (173)29
91.180.54101 (95)181 (181)31


Both approaches studied in this work allow the acquisition of whole-body 3D head-to-toe water/fat images at different spatial resolutions in a single run through the magnet. They are both based on chemical shift information, which shows clear advantages over techniques that estimate the fat signal based on a specific T1- or T2- image contrast and image postprocessing (1) by avoiding partial volume effects and potential contrast ambiguities. However, the presented results make improvement of the image quality desirable. The magnetization-prepared approach has problems in regions of high susceptibility gradients (neck, shoulder). Signal components of fatty tissue contribute to the water image (see arrows in Fig. 3a), whereas the fat image shows some signal dropouts in that critical region. Although the use of a patient-independent off-resonance Larmor frequency lookup table improved the image quality considerably, as was proved by scans performed without this correction (not shown here), this measure was not sufficient to compensate for the off-resonance in some of the volunteers. Compensation may be even more difficult to obtain when patients are examined who show larger size variation than the male volunteers in this study. Improved performance might be attainable using higher resolution of the 1D off-resonance correction table or nonlinear matching to each patient's anatomy. Even better performance may be obtainable by applying patient-specific correction, e.g., based on repeated measurements of the resonance frequency interleaved into the data acquisition (17).

The chemical shift encoding approach is more robust against the patient-induced off-resonance, since B0 inhomogeneity is taken into account in the iterative algorithm. However, some deficiencies in the image quality are seen here, also, in the form of slight dark-band artifacts occurring with a z-periodicity of L near the body's periphery (shoulder, arm). These artifacts are already encoded into the images of the individual echoes and become more prominent with increasing TE. They are caused by B0- and B1- inhomogeneities of the scanner (as opposed to patient-induced susceptibility), which act as additional signal encoding for moving objects. Each voxel, traveling at the table velocity, experiences this inhomogeneity during k-space traversal and acquires an extra signal phase that varies among the individual lines in the hybrid space. The resulting artifacts are prominent in those regions where the central phase-encoding steps are measured, that is, where small phase errors result in strong signal smearing along the phase-encoding directions after FT. The magnetization prepared approach is less sensitive to this effect because a smaller value can be chosen for TE.

In order to minimize the propagation of these banding artifacts into the solution of the iterative water/fat separation, a short value of the longest TE is desirable. On the other hand, to minimize the noise propagation from the images of the individual echoes into the water/fat solution, the fitting problem has to be sufficiently well posed, which requires sufficiently long encoding times. Based on the analysis given in Ref.18, the SNR gain of the water/fat solution for the encoding scheme of protocol II corresponds statistically to an equivalent of 2.75 averages. This represents a slight but acceptable loss, considering the three measured images used. Consequently, one has to find an appropriate compromise between the longest TE and the pixel bandwidth to weigh SNR loss against artifacts in the water/fat images.

The measures for body height, weight, and BMI showed a nice agreement with their conventionally-obtained values. The global water/fat ratio is a simple, but interesting parameter and is directly related to the whole-body water/fat composition. More costly from a data-processing point of view is the determination of the intraperitoneal/extraperitoneal fat ratio. But this parameter is important as it is considered to be of high diagnostic value (19); e.g., to characterize patient specific risk factors. Further work is necessary to improve data analysis. The water/fat signals determined in this study are still biased by T1/T2 effects, which makes true quantification difficult. Also, more appropriate qualifiers have to be developed to extract as much diagnostic information as possible from this type of fast scan, which could easily be added to other clinical MRI exams.

The two water/fat selective continuously moving table approaches described in this work can find a number of useful applications. The different spatial resolution scales can be exploited in different directions. High resolution could be important, for example, to address perilesional fat sparing, which can serve as an indicator for tumors and their metastases (20). Low resolution is interesting for fast fat screening to investigate obesity related pathology and to study the complex processes of fat deposition, which so far is poorly understood (1).

In conclusion, this study showed the feasibility of 3D whole-body water/fat resolved continuously moving table MRI. The three-point chemical shift encoding approach was found to be more efficient than the magnetization prepared technique. Image quality is promising, but further optimizations are necessary. This technology has the potential to allow whole-body head-to-toe water/fat resolved imaging with high patient comfort at adjustable spatial resolution and speed.